University of Southern Denmark

Notes on Chern-Simons perturbation theory

Wernli, Konstantin

Published in:
Reviews in Mathematical Physics

DOI:
10.1142/S0129055X22300035

Publication date:
2022

Document version:
Accepted manuscript

Citation for pulished version (APA):

Wernli, K. (2022). Notes on Chern-Simons perturbation theory. Reviews in Mathematical Physics, 34(3), Article
2230003. [Link]

Go to publication entry in University of Southern Denmark's Research Portal

Terms of use
This work is brought to you by the University of Southern Denmark.
Unless otherwise specified it has been shared according to the terms for self-archiving.
If no other license is stated, these terms apply:

• You may download this work for personal use only.

• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying this open access version
If you believe that this document breaches copyright please contact us providing details and we will investigate your claim.
Please direct all enquiries to puresupport@[Link]

Download date: 28. Jan. 2026

Notes on Chern-Simons Perturbation Theory

Konstantin Wernli

October 29, 2021

 Abstract

We give a detailed introduction to the classical Chern-Simons gauge theory, including the
mathematical preliminaries. We then explain the perturbative quantization of gauge theories
via the Batalin-Vilkovisky (BV) formalism. We then define the perturbative Chern-Simons
partition function at any (possibly non-acylic) reference flat connection using the BV for-
malism, using a Riemannian metric for gauge fixing. We show that it exhibits an anomaly
known as the “framing anomaly” when the Riemannian metric is changed, that is, it fails to
be gauge invariant. We explain how one can deal with this anomaly to obtain a topological
invariant of framed manifolds.
Contents

Preface i

Introduction 1

1 Classical Chern-Simons Theory 10

1.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.1.1 Vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.1.2 Principal bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.1.3 Connections on vector bundles . . . . . . . . . . . . . . . . . . . . . . 19

1.1.4 Connections on principal bundles . . . . . . . . . . . . . . . . . . . . 21

1.2 Chern-Simons action functional . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.2.1 The Chern-Simons 3-form . . . . . . . . . . . . . . . . . . . . . . . . 27

1.2.2 The action functional . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.3 Critical Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1.4 The representation variety . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

1.4.1 The horizontal distribution of a connection . . . . . . . . . . . . . . . 33

1.4.2 Parallel transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

i
 1.4.3 The isomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

1.4.4 Topology and smooth structure . . . . . . . . . . . . . . . . . . . . . 37

2 Perturbative Quantization 39

2.1 Asymptotics of oscillatory integrals . . . . . . . . . . . . . . . . . . . . . . . 40

2.1.1 Stationary phase formula . . . . . . . . . . . . . . . . . . . . . . . . . 40

2.1.2 Higher order corrections . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.1.3 Feynman diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2.1.4 Feynman graphs and rules . . . . . . . . . . . . . . . . . . . . . . . . 49

2.2 Oscillatory integrals with degenerate phase functions and Faddeev-Popov method 50

2.2.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.2.2 Gauge fixings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2.2.3 Super vector spaces and Berezin integrals . . . . . . . . . . . . . . . . 53

2.2.4 The Faddeev-Popov action . . . . . . . . . . . . . . . . . . . . . . . . 55

2.2.5 Stationary phase method for Faddeev-Popov integrals . . . . . . . . . 57

2.2.6 Example: Yang-Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

2.2.7 Example: Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

2.3 BRST symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

2.3.1 BRST operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

2.3.2 BRST cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

2.3.3 Gauge fixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

2.4 Batalin-Vilkovisky formalism and effective actions . . . . . . . . . . . . . . . 66

2.4.1 Odd symplectic vector spaces . . . . . . . . . . . . . . . . . . . . . . 66

ii
 2.4.2 Odd Poisson bracket and BV algebra . . . . . . . . . . . . . . . . . . 67

2.4.3 BV integral and gauge fixing independence . . . . . . . . . . . . . . . 68

2.4.4 Master equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

2.4.5 Faddeev-Popov and BRST solutions to Classical Master Equation . . 70

2.4.6 BV pushforward and effective actions . . . . . . . . . . . . . . . . . . 70

3 Perturbative Quantization of Chern-Simons Theory 73

3.1 BV Chern-Simons theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.1.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.1.2 BRST operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.1.3 BV formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3.1.4 The Quantum Master Equation . . . . . . . . . . . . . . . . . . . . . 76

3.2 Gauge fixing using background fields . . . . . . . . . . . . . . . . . . . . . . 77

3.2.1 Fixing a background field . . . . . . . . . . . . . . . . . . . . . . . . 77

3.2.2 The gauge fixing Lagrangian . . . . . . . . . . . . . . . . . . . . . . . 77

3.2.3 Inverting the quadratic term . . . . . . . . . . . . . . . . . . . . . . . 80

3.2.4 The propagator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

3.3 The effective action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.3.1 Feynman graphs and rules . . . . . . . . . . . . . . . . . . . . . . . . 84

3.3.2 The superdeterminant and the Ray-Singer torsion . . . . . . . . . . . 88

3.3.3 The phase factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

3.4 The Quantum Master Equation, again . . . . . . . . . . . . . . . . . . . . . 91

3.4.1 The Laplacian on the space of residual fields . . . . . . . . . . . . . . 92

iii
 3.4.2 Proof of Quantum Master Equation . . . . . . . . . . . . . . . . . . . 93

3.5 Gauge invariance and the framing anomaly . . . . . . . . . . . . . . . . . . . 97

3.5.1 Dependence of Zf ree on the gauge-fixing metric . . . . . . . . . . . . 97

3.5.2 Gauge dependence of Zpert . . . . . . . . . . . . . . . . . . . . . . . . 99

3.5.3 The framing anomaly in the literaure . . . . . . . . . . . . . . . . . . 101

iv
Introduction

In this article we will define the perturbative Chern-Simons partition function in the Batalin-
Vilkovisky formalism, and explain why it fails at first sight to be gauge invariant, and how
to fix this and obtain invariants of (framed) 3-dimensional manifolds from Chern-Simons
theory. We will explain what this means in more detail below. The ultimate goal of this
study of the perturbative quantization of Chern-Simons theory is a better mathematical
understanding of the Feynman path integral [Fey42] in the domain of quantum field theory.
Let us briefly introduce this notion.

The Feynman path integral

Let us take the simplifying viewpoint that physics is about the prediction of values of ob-
servables, i.e. numbers attached to outcome of an experiment. The principle of least action
provides the following recipe for this computation in classical physics. By a d-dimensional
classical field theory, we mean an assignment that assigns to d-dimensional manifolds1 M
the following data:

• A topological space called “space of fields” FM ,

• A continuous function SM : FM → R called “action functional”.

1
Often these manifolds are subject to additional conditions (such as compactness) or required to be
equipped with extra structure (such as an orientation or a metric).

1
The manifold M is called the “spacetime” manifold. It is a basic requirement from physics
that the assignment M 7→ (FM , SM ) be local.2 The principle of least action then states that
the physical field configuration satisfies

δS[φ0 ] = 0, (1)

the Euler-Lagrange equation 3 (here δ denotes the variational derivative). Supposing for a
moment a unique solution φ0 to this equation, the value of any observable O : FM → R can
be computed as O(φ0 ) (that is, if we were to measure the observable O, the outcome would
be O(φ0 )). Let us briefly look at two easy examples.
A common example is free scalar field theory. Here M is a Riemannian (or Lorentzian)
manifold, for example M = R4 with standard Euclidean (or Minkowski) metric. The space
of fields is FM = C ∞ (M, R) and the action functional is given by
Z Z
1 2 2 2 1
SM = (||dφ|| + m φ ) dvolg = (φ∆g φ + m2 φ2 ) dvolg , (2)
2 M 2 M

where || · || is the norm induced by the metric on the cotangent bundle, ∆g the Laplacian
induced by the metric, and finally, dvolg denotes the volume form of the metric. Variation
of this action functional leads to the Helmholtz equation

∆φ + m2 φ = 0

and the analysis proceeds by looking for eigenfunctions of the Laplace operator. Possible
R
observables are for instance Ox0 (φ) = φ(x0 ) for a point x0 ∈ M or Oη (φ) = ηφ dvolg for a
test function η ∈ Cc∞ (M ).
Another example is classical mechanics. Here, the “space-time” is just a (time) interval I
and the space of fields is C ∞ (I, N ) where N is a Riemannian manifold modeling space. The
action functional is Z
1
SI = m||γ̇(t)||2 − V (γ(t)) dt, (3)
I 2
2
There are different mathematical manifestations of this physical concept. One possibility is to ask that
is that FM be the space of sections of a bundle E over M , and that SM is the integral over M of a local
Lagrangian density: A density-valued function L on the k-th jet bundle J k (E) of E (the minimal such k is
called order of the theory, remarkably, most physical theories are either of first or second order).
3
Of course, this equation is only necessary for φ0 to be an actual minimizer of the action, but nowhere
near sufficient: usual questions about critical points apply, with the added complication of FM usually being
infinite-dimensional.

2
where V : N → R is called the potential. The Euler-Lagrange equation is Newton’s Law

mγ̈ = (∇V )(γ).

Possible observables include the position or velocity at some time t0 : Ot0 (γ) = γ(t0 ), Ot0 0 (γ) =
||γ 0 (t0 )|| and so on.
We now want to pass from classical to quantum physics. This is a major conceptual leap
that we cannot possibly do justice here.4 A main feature of quantum physics is that one can
no longer predict values of observables with certainty, but only with certain probabilities.
We are taking here (a very much simplified version of) Feynman’s approach to quantum
field theory [Fey42; Fey49a; Fey49b; Fey50]. It dictates that the expectation value of an
observable can be computed as
Z
1 i
hOiM = O(φ)e ~ SM [φ] Dφ, (4)
ZM FM

where the partition function Z is given by

Z
i
ZM = e ~ SM [φ] Dφ. (5)
FM

Integrals (4) and (5) are examples of what is known as Feynman path integrals (or also
functional integrals). We will not delve further into their physical origins and interpretations
(which are very elegant and interesting), but rather investigate the mathematical nature of
these integrals. For a rigorous definition of an integral, a measure is required. But, the
definition of sensible5 measures on spaces of field seems not possible in general, with the
remarkable exception of the two examples above, see [GJ87] for a deeper discussion of this
fact and further references.
Let us for the moment focus our discussion on the partition function ZM and the assignment

Z : M 7→ ZM . (6)

If this cannot be defined via an actual integral, the question is how else it can be rigorously
mathematically defined? The continued success of the use of functional integration tech-
niques suggests to mimick some properties of integrals in assignment (6). Two ideas as to
which properties to use are prominent in the mathematical community:
4
The precise nature of the relation between classical and quantum physics has been the subject of research
for over a century, and still not been fully understood.
5
For instance, countably additive Borel.

3
 i) Try to implement Fubini’s theorem in M 7→ ZM ,

ii) Try to implement the asymptotics6 of oscillatory integrals in M 7→ ZM .

The first approach leads to the idea of functorial quantum field theory (or FQFT). The
second approach leads to what we call perturbative quantization in this lecture.7 We briefly
illuminate both approaches (the second one will be discussed in detail later in the text).

Functorial Quantum Field Theory

Let us briefly explain how supposing that the path integral satisfies Fubini’s theorem can
be interpreted as Functorial QFT. The basic idea is that the locality assumption of field
theories allows to cut the spacetime into pieces. Namely, for manifolds with boundary M
one can define a space of boundary fields F∂M with a surjective map π : FM → F∂M which
in the simplest case is just restriction of the fields to the boundary.8 If M is a manifold
that can be represented as the union of two manifolds M1 and M2 joined along a common
boundary Σ: M = M1 ∪Σ M2 (for instance, S 2 = D2 ∪S 1 D2 can be glued from two disks
along their common boundary circle S 1 ) then, locality dictates that

FM = FM1 ×FΣ FM2 = {(φ1 , φ2 ) ∈ FM1 × FM2 |π1 (φ1 ) = π2 (φ2 )} (7)

and
SM [(φ1 , φ2 )] = SM1 [φ1 ] + SM2 [φ2 ]. (8)
6
The physical constant ~ is very small and it makes sense to ask about the properties of the semiclassical
limit ~ → 0.
7
Perturbation theory in physics is usually performed in the coupling constant, and not in ~ (which is
usually set to 1). So one could argue that we are speaking about the semiclassical approximation, rather
than perturbative quantization. However, the two approaches differ only by an overall rescaling, and are
mathematically equivalent.
8
Typically this is only the case in first order theories, in theories of higher order one needs to also consider
normal derivatives.

4
In other words, FM is a fiber bundle over FΣ with fiber (FM )b = π1−1 (b) × π2−1 (b). Now,
applying a formal Fubini theorem9 implies that we can factorize the integral (5) as
Z Z Z 
i i
S [φ] S [φ]
ZM = e ~ M Dφ = e ~ M Dφ Db
FM FΣ (FM )b
Z Z ! Z !
i i
= e ~ SM1 [φ1 ] Dφ1 e ~ SM2 [φ2 ] Dφ2 Db (9)
FΣ π1−1 (b)⊂(FM1 ) π2−1 (b)FM2

(here we use integration along fibers in the first line and Fubini theorem in the second). Intro-
R
duce the vector space10 HΣ = Fun(FΣ ) with the (formal) pairing hf1 , f2 iHΣ = FΣ f1 (b)f2 (b)Db,
then we can rewrite Equation (9) as

ZM = hZM1 , ZM2 iHΣ . (10)

From this formal considerations, one extracts the following definition (usually attributed to
Atiyah [Ati88] and Segal [Seg88]).

Definition. A functorial QFT, denoted Z, associates

• To every d − 1-dimensional manifold Σ a vector space Z(Σ) =: HΣ with an inner

product h·, ·iH

• To a d-dimensional manifold M with boundary a vector Z(M ) ∈ H∂M

such that

• H∅ = R

• If M = M1 ∪Σ M2 , then
ZM = hZM1 , ZM2 iΣ . (11)

The adjective functorial stems from the fact that such an assignment can be made into a
functor11 from a cobordism category (where objects are closed d − 1 dimensional manifolds
9
Together with its generalized cousin for integration along fibers.
10
Here Fun denotes some space of functions whose precise nature is irrelevant to the present heuristic
discussion.
11
Mathematicians love functors.

5
Σ. and morphisms from Σ1 to Σ2 are d-dimensional manifolds with boundary Σ1 t Σ2 ) to
the category of vector spaces and linear maps. Functors out of cobordism categories became
of great interest to the mathematical community (and, partly, also physicists) thanks to this
interpretation of the path integral, and over the last thirty years a considerable amount of
research has gone into this area, and its extension to higher codimension (i.e. manifolds
with corners etc.) and higher categories. Reviews include [Koc03], [Tur94]. In some sense
the study of functors out of cobordism categories culminates in the “cobordism hypothesis”
- very roughly, it says that functors that can be extended to maximal codimension (i.e. to
points) are entirely determined by their values on points. It was proposed in [BD95] and
recently proven by Lurie in [Lur09]. Further references include [Kap10], [Fre12], [AF17]. The
subject also has applications on condensed matter physics, see for instance [FH16], [KT17].

Perturbative Quantization

We will be very brief here since this issue is discussed in great detail later. Another obser-
vation about Feynman integrals is that since SM is real, the function exp(i/~SM ) oscillates
very wildly as ~ → 0, with exception of the critical points.12 For finite-dimensional integrals,
this leads to the well-known principle of stationary phase. Suppose F is a finite-dimensional
manifold (with a reference density µ) and S : F → R has isolated non-degenerate critical
points. Then as ~ → 0, the integral
Z
I[~] = exp(i/~S) µ (12)
M

is concentrated in a neighborhood of the critical points of S. Around a critical point one

can expand S in a Taylor series and then compute integrals of terms explicitly using Fresnel
integrals.13 In particular, there is an explicit expression in form of a formal power series for
the asymptotic behavior of I[~]
iπ 00
1 X
i/~S(x0 ) e 4 sign S (x0 )
I[~] '~→0 e (1 + O(~)) (13)
2π~dim F/2 det 1/2 S 00 (x0 )
x0 ∈Crit(S)

12
Sometimes this is rephrased in Euclidean setting where one considers exp(−S/~) as using method of
steepest descent. The two approaches essentially produce equivalent answers to our question of understanding
the path integral in the perturbative sense.
13
The equivalent of Gaussian integrals for complex exponents.

6
which uses only the Taylor expansion of S at the critical point and the inverse of its quadratic
part at the critical point (and its determinant). The terms in this formal power series of
order14 O(~) can be conveniently labeled by diagrams - later we will identify them as Feynman
diagrams. This has the major advantage that it can be generalized to infinite dimensions, if
we can make sense of the Taylor expansion, the determinant of the inverse, and so on (but
this has proven to be a lot simpler than finding appropriate measures). We will discuss all
of these issues in Section 2.

Why study Chern-Simons Theory?

The Chern-Simons gauge theory is defined for 3-manifolds M , for a precise definition of FM
and SM we refer to Section 1.2. It is a topological theory, meaning that its definition does
not involve the data of a metric (Riemannian or Lorentzian) on M .
To make a long story short, the answer to the question asked in the title is that Chern-Simons
theory has been studied using a variety of approaches and viewpoints. In this sense, it is one
of the best opportunities to understand the Feynman path integral, because so many answers
are available that one can compare to and use in the task. However, the question about the
precise relationships between the different results still remains wide open. A better under-
standing of these relationships will deepen our understanding of the concept of quantization
itself.
To be slightly more precise, after the seminal paper of Witten [Wit89], interest in Chern-
Simons theory in the mathematical physics community exploded, making it one of the most
well-studied field theories at both classical and quantum level (in some sense the drosophila
melanogaster of quantum field theory). An exhaustive review of the literature is next to
impossible, and we restrict ourselves to mentioning a few results.
It was Witten who argued that Chern-Simons theory was linked closely to knot and 3-
manifold invariants in [Wit89]. To be more precise, he argued - using holomorphic quantiza-
tion of the reduced phase space - that expectation values of Wilson loop observables are given
by the Jones polynomial, which was discovered by Jones when investigating von Neumann
algebras [Jon85]. Around the same time, Fröhlich and King [FK89] showed that also the
perturbative quantization of Chern-Simons theory on R3 with Wilson lines leads to knot in-
14 1
Ignoring the overall constant 2π~dim F /2

7
variants, via the Khnizhnik-Zamolodchikov connection (these invariants later became known
as the Kontsevich integral [Kon93]). From this moment on it was clear that Chern-Simons
theory was intimately connected with the vast subjects of knot and 3-manifold invariants,
and conformal field theory.
Shortly after Witten, Reshetikhin and Turaev [RT91] defined a TQFT that led to answers
similar to Witten’s, and this TQFT is widely considered as the correct non-perturbative
quantization of Chern-Simons theory, even though - to the best of the author’s knowledge -
there is no conclusive proof of a mathematical formulation of this statement.
The perturbative quantization of Chern-Simons was considered in various formulations and
guises, starting in the more physics-oriented literature with [GMM89]. Shortly thereafter,
Axelrod-Singer [AS91],[AS94] and - in a different way - Kontsevich [Kon94] showed that one
can obtain 3-manifold invariants from the perturbative quantization of Chern-Simons theory
on 3-manifolds (subject to certain conditions). The precise link between these invariants and
the “non-perturbative” ones defined by Reshetikhin and Turaev is still unclear (and one of the
main motivations for this lecture). The perturbative approach after Axelrod-Singer was de-
veloped further by Bott, Cattaneo, and Mnëv in the papers [BC98], [BC99], [CF00],[CM08],
the main references for this article. Casting Chern-Simons theory in the BV formalism (as
in [CM08]) has the advantage that one can define the perturbation around arbitrary flat
connections.15 This is an important step towards being able to compare the Chern-Simons
perturbation theory with the “non-perturbative” invariants defined by Reshetikhin and Tu-
arev.
Let us also mention that Chern-Simons theory has been studied from the viewpoint of geo-
metric quantization [APW91] and conformal field theory (see e.g. [AU15]). It is the author’s
belief that a thorough understanding of the relationships between these different approaches
to the quantum Chern-Simons theory will elucidate the evasive mathematical underpinnings
behind the Feynman path integral, and in the process contribute to a deeper understanding
of the involved mathematical subjects.
15
This aspect was not pursued in detail in [CM08], which was one of the author’s main motivations for
writing this article.

8
Outline of the text

We finish the introduction by giving a brief outline of the text.

In Section 1 we discuss the classical Chern-Simons theory. In Section 1.1, we explain the
mathematical preliminaries: Vector bundles, principal bundles, connections and gauge trans-
formations, before looking in detail at the classical Chern-Simons action functional in Section
1.2 and its critical points, the flat connections, in Section 1.3. In Section 1.4 we also discuss
the moduli space of critical points, i.e. the space of flat connections up to gauge transfor-
mations.

In Section 2 we explain the perturbative quantization of gauge theories. We start by review-

ing the asymptotics of oscillatory integrals in Section 2.1, and how they lead to Feynman
graphs and rules. We then proceed to analyze oscillatory integrals with degenerate phase
functions and explain the gauge-fixing formalism of Faddeev and Popov in Section 2.2. In
particular, we see that there is a choice involved when computing such asymptotics, and we
discuss the independence of these choices. We then briefly comment on the BRST symmetry
in Section 2.3 before finally turning our attention to the Batalin-Vilkovisky formalism in
Section 2.4. In particular, we explain the BV Laplacian ∆, and show that in the finite-
dimensional case the analog of the partition function gives a class in the ∆-cohomology that
is independent of the choice of gauge-fixing.

In Section 3, we analyze the perturbative quantization of the Chern-Simons gauge theory.

We first explain how to cast it in the BV setting in Section 3.1. In Section 3.2, we explain how
to gauge fix the theory using a choice of Riemannian metric. We then proceed to define the
effective action on zero modes in Section 3.3. We show that the effective action satisfies the
quantum master equation, i.e. its exponential is ∆-closed, in Section 3.4. This means that
the partition function, the exponential of the effective action, defines an element in the BV
cohomology. In the final Section 3.5, we show that this cohomology class in not independent
of the gauge-fixing Riemannian metric, this is typically called the “framing anomaly”. We
also explain how to fix the anomaly to obtain an invariant of framed 3-manifolds from the
perturbative partition function of Chern-Simons theory.

9
Section 1

Classical Chern-Simons Theory

In this Section we introduce the classical Chern-Simons action functional on closed manifolds
and study its symmetries and critical points. There are some slightly subtle geometric effects
in the definition of that action functional, related to the trivializability of bundles. For that
reason we review the concepts of vector bundles, principal bundles and connections. More
details can be found in any textbook on gauge theory, e.g. the book by Taubes [Tau11] or
the lecture notes by Baum [Bau14].

1.1 Preliminaries

In this section we review the basic notions of gauge theory: vector bundles, principal bundles,
and connections.

1.1.1 Vector bundles

We start with the definition of a vector bundle.

Definition 1.1.1 (Vector bundle). Let M be a manifold and k ∈ {R, C}. A rank n k-
vector bundle over M is a pair (E, π), where E is a manifold and π : E → M is a surjective
submersion, such that there is an open cover U = {Uα }α∈A of M satisfying

10
 i) The cover U trivializes E, that is, for every α ∈ A there exists a diffeomorphism
ψα : π −1 (Uα ) → Uα × k n such that
ψα
π −1 (Uα ) Uα × k n
π π1 (1.1)

Uα
commutes,

ii) For all u ∈ M the preimage π −1 (u) is a k-vector space and for all α ∈ A with u ∈ Uα ,
the map
ψα π −1 (u)
: π −1 (u) → {u} × k n (1.2)
is an isomorphism of k-vector spaces.

Let us introduce some terminology. M is called the base (or base space) of the vector bundle.
E is called the total space, and π the projection. For u ∈ M , π −1 (u) is called the fiber (of
E) over u, and denoted Eu . The pair (Uα , ψα ) is called a local trivialization and U is called
a trivializing cover. For α, β ∈ A, let Uαβ = Uα ∩ Uβ . By diagram (1.1) and (1.2), the maps1

g̃αβ = ψβ ◦ ψα−1 : Uαβ × k n → Uαβ × k n (1.3)

satisfy g̃αβ (u, v) = (u, gαβ (u)v), where gαβ (u) ∈ GLn (k). The corresponding maps

gαβ : Uαβ → GLn (k) (1.4)

are called the gluing maps. By construction, they satisfy, for all α, β, γ ∈ A

gαα (u) = idkn , ∀u ∈ Uα (1.5a)

gαβ (u) = gβα (u)−1 , ∀u ∈ Uαβ (1.5b)
gβγ (u)gαβ (u) = gαγ (u), ∀u ∈ Uαβγ = Uα ∩ Uβ ∩ Uγ . (1.5c)

Definition 1.1.2 (Vector bundle morphisms). If E, F are vector bundles over M then a
vector bundle morphism is a smooth map Ψ : E → F such that the diagram
Ψ
E F
π π0 (1.6)
M
1
Often, in the literature one finds opposite convention for the indices. However, we find this intuitive
because it is the transition map from α to β.

11
commutes and Ψ Eu
=: Ψu : Eu → Fu is linear. A vector bundle isomorphism is a vector
bundle morphism which is also a diffeomorphism. The set of vector bundle morphisms from
E to F is denoted Hom(E, F ).

Remark 1.1.3. Given two vector bundles E and F over M , we can always find a cover of M
that trivializes both. Namely, given a trivializing cover {Uα }α∈A of E and {Vβ }β∈B of F , the
cover {Uα ∩ Vβ }(α,β)∈A×B is a trivializing cover of both E and F .
Exercise 1. Suppose the rank of E is n and the rank of F is m. Prove that a vector bundle
morphism Ψ : E → F is given by a collection of maps Ψα : Uα → Hom(k n , k m ) such that

F E
Ψβ = gαβ Ψα gβα . (1.7)

Example 1.1.4. i) A vector space is a vector bundle over a point.

ii) For every manifold M , the tangent bundle T M is a vector bundle over M . The transition
maps of the tangent bundle T M can be computed in the following way. Let (Uα , ϕα ) be
an atlas of M . Then dϕαβ (u) : Uαβ → GLn (R) are the transition maps of T M .

iii) For any manifold M and natural number n there is the trivial rank n vector bundle over
k, simply given by the direct product M × k n with the canonical projection to M . This
bundle is often denoted k n .

It is an important fact that the vector bundle is entirely determined up to isomorphism by

its trivializing cover and the gluing maps.

Proposition 1.1.5. Given a cover U = {Uα }α of a manifold M and a family of smooth

maps gαβ : Uαβ → GLn (k) satisfying (1.5), there exists a unique (up to isomorphism) vector
bundle π : E → M with trivializing cover U and gluing maps gαβ .

Proof. Existence: We can assume that each Uα is a contained in a domain of a chart of M

(otherwise, cover each Uα by charts Vαβ and note that the transition functions restricted to
each Vαβ still satisfy (1.5)). First, construct the fiber over u by
!
a
Eu := k / ∼= ({α ∈ A|u ∈ Uα } × k n ) / ∼
n
(1.8)
α∈A,u∈Uα

12
where (α, v) ∼ (β, w) if gαβ (u)v = w. This is an equivalence relation since gαβ satisfy (1.5).
`
Then, let E := u∈M Eu and ψα [(u, α, v)] = (u, v). The projection π : E → U is given by
π(Eu ) = {u}. Since Uα is contained in a chart, composition with this chart yields a chart of
E. It is easily checked that this is indeed a smooth atlas.
Uniqueness: It is enough to show that two vector bundle with the same trivializing cover
and gluing maps are isomorphic. Let E,F be such vector bundles. Then, we construct the
isomorphism over Uα by the diagram
E
ψα F )−1
(ψα
Ψα : π −1 (Uα ) Uα × k n (π 0 )−1 (Uα )
π π0

Uα

Since the gluing maps are the same, the maps Ψα and Ψβ agree on Uαβ : Indeed we have

Ψβ = (ψβF )−1 ◦ ψβE

F
= (g̃αβ ◦ ψαF )−1 ◦ (g̃αβ
E
◦ ψαE )
= (ψαF )−1 ◦ (g̃αβ
F −1 E
) ◦ g̃αβ ◦ ψαE = Ψα ,

F −1 E
since (g̃αβ ) ◦ g̃αβ = idUαβ ×kn . Using Exercise 1, we conclude that the maps Ψα assemble
into a vector bundle isomorphism Ψ : E → F .

This central fact will often help us define vector bundles via trivializing covers and gluing
maps. Given a manifold M , we can define a rank n vector bundle E over k by specifying a
trivializing cover U and transition maps gαβ : Uαβ → GLn (k), and we write E = (U, gαβ ) for
this vector bundle.

Example 1.1.6. Consider the circle S 1 = R/Z with the open cover U1 = (0, 1), U2 =
(1/2, 3/2). Then the Möbius strip is the line bundle with transition function g12 : U12 →
GL1 (R) = R× given by 
1 x ∈ (1/2, 1)
g12 (x) =
−1 x ∈ (1, 3/2)

In some sense, this is the only interesting line bundle over the circle:
Exercise 2. a) Show the tangent bundle of S 1 is a trivial line bundle: T S 1 ∼
= S 1 × R.

13
b) Show that any line bundle over S 1 is isomorphic to either the Möbius strip or the trivial
line bundle.

Definition 1.1.7 (Section). A section of a vector bundle π : E → M is a smooth map

σ : M → E such that π ◦ σ = idM .

The set of sections of E is denoted Γ(M, E) or simply Γ(E) when no confusion is possible.
Note that since Ex is a vector space for all x ∈ M , we can naturally add sections and multiply
them by scalars: For σ, σ1 , σ2 ∈ Γ(E) and λ ∈ k we define

(σ1 + σ2 )(x) = σ1 (x) + σ2 (x), (λσ)(x) = λσ(x). (1.9)

Thus, Γ(E) is a k-vector space. We can also multiply section by smooth functions: If
f ∈ C ∞ (M, k) and σ ∈ Γ(E) we define

(f σ)(x) = f (x)σ(x).

This shows that Γ(E) is in fact a module over the ring C ∞ (M, k).

Example 1.1.8. a) A section of a trivial bundle M × k n is given by σ(x) = (x, f (x)), where
f : M → k n is a smooth map. Thus, Γ(M, k n ) ∼
= C ∞ (M, k n ).

b) Over a trivializing cover U = {Uα }α∈ A , a section is given by smooth functions σα : Uα →

k n satisfying
σβ (x) = gαβ (x)σα (x). (1.10)

c) A section of the tangent bundle T M is called a vector field.

Exercise 3. Show that a line bundle that admits a nowhere vanishing section is isomorphic
to a trivial bundle.

The natural constructions on vector spaces, such as dualizing, direct sums and tensor prod-
ucts, carry over to vector bundles. Here the description in terms of transition functions
comes in handy.

Definition 1.1.9. Let E = (U, gαβ ) and F = (U, hαβ ) be two vector bundles over the same
trivializing cover. Then we define the following bundles:

14
 i) The dual bundle E ∗ by
∗ −1
E ∗ = (U, (gαβ ) ) (1.11)

ii) The direct sum E ⊕ F by

E ⊕ F = (U, gαβ ⊕ hαβ ) (1.12)

iii) The tensor product E ⊗ F by

E ⊗ F = (U, gαβ ⊗ hαβ ) (1.13)

Vk
iv) The symmetric and exterior powers Symk E and E by

Symk E = (U, Symk gαβ ) (1.14)

k k
!
^ ^
E = U, gαβ . (1.15)

v) The determinant line det E by

rk(E)
^
det E = E. (1.16)

Example 1.1.10. a) The dual of the tangent bundle T M is called the cotangent bundle
and denoted (T M )∗ = T ∗ M .
Vk
b) Sections of T ∗ M are called differential k-forms on M and denoted Ωk (M ). We also
denote Ω• (M ) = dim
L M k
k=0 Ω (M ).

Vk
c) If E is a vector bundle, then sections of T ∗ M ⊗ E are called differential k-forms with
values in E, and denoted Ωk (M, E).

d) Sections of E ∗ ⊗ F are the same as vector bundle morphisms E → F :

Γ(E ∗ ⊗ F ) ∼
= Hom(E, F )

(Exercise!)

15
1.1.2 Principal bundles

We begin with a definition.

Definition 1.1.11 (Principal bundle). Let G be a Lie group and M be a manifold. A

principal G-bundle is a triple (P, π, M ) such that

i) the mapπ : P → M is a smooth submersion,

ii) there is a right action P × G → P such that π is G-invariant (that is, π(pg) = π(p)),

iii) the right action is free and transitive when restricted to fibers π −1 (u) for all u ∈ M ,

iv) There exists a trivializing cover U = {Uα }α∈A , that is, a cover of M with the property
that for every α ∈ A there exists a diffeomorphism Ψα : π −1 (Uα ) → Uα × G such that
Ψα
π −1 (Uα ) Uα × G
π π1 (1.17)

Uα

commutes and Ψα (pg) = Ψα (p)g.2

Again, we call M the base and P the total space of the bundle. The Lie group G is called
the structure group of the bundle.

Definition 1.1.12 (Morphism of principal bundles). A morphism of principal G-bundles P

and P 0 is a smooth map f : (P, π) → (P 0 , π 0 ) that commutes both with the right G-action
and the projections, that is

f (pg) = f (p)g (1.18)

π(f (p)) = π(p), (1.19)

and an isomorphism of principal G-bundles is a morphism which is also a diffeomorphism.

Similarly to the case of vector bundles we define

g̃αβ = Ψβ ◦ Ψ−1
α : Uαβ × G → Uαβ × G (1.20)
2
On the right hand side of this equation G acts trivially on Uα and by right multiplication on itself.

16
which are given by
g̃αβ (u, h) = (u, gαβ (u)h) (1.21)

since the trivializations commute with the projections. The maps

gαβ : Uαβ → G (1.22)

are called the transition or gluing maps. Note that the transition functions act from the left
and commute with the right G-action. They satisfy

gαα (u) = 1G (1.23a)

−1
gαβ (u) = gβα (u) (1.23b)
gαγ (u) = gβγ (u)gαβ (u) (1.23c)

We also have a similar proposition:

Proposition 1.1.13. Let U = {Uα }α∈A be a cover of M and suppose gαβ : Uαβ → G satisfy
(1.23). Then there exists a principal G-bundle P over M with trivializing cover U and gluing
maps gαβ , and this bundle is unique up to isomorphism.

Even though vector bundles and principal bundles are different objects, in some sense they
are like two sides of the same coin. This is explained by the following fundamental example:

Example 1.1.14. If V is a vector space, then a frame of V is an ordered basis e =

(e1 , . . . , en ). The set of frames is denoted by Fr(V ). Let π : E → M be a rank n k-vector
bundle. Then, the frame bundle Fr(E) of E is the smooth manifold Fr(E) = tx∈M Fr(Ex ).
This manifold has a natural projection π : Fr(E) → M . We can define a right GLn (k) action
on Fr(E) in the following way:

e · g = g −1 (e1 , . . . , en ) (1.24)

This gives Fr(E) the structure of a principal GLn (k)-bundle. We can find a trivializing cover
as follows. Let {Uα }α be a trivializing cover for E. Over a trivializing chart ψα : π −1 (Uα ) →
Uα × k n , with ψα = (π, Aα ), then a trivialization of Fr(E) is given by

ψ α (e) = (π(e), Aα (e)).

17
Here Aα (e) = (Aα e1 , . . . Aα en ) ∈ GLn (k). Now, one can check that this principal GLn (k)-
bundle has the same gluing maps gαβ = Aβ A−1
α :

Aβ (e) = Aβ (Aα A−1

α )(e) = gαβ Aα (e),

hence gαβ are the transition functions of P since they satisfy the defining equation

ψ β ◦ ψ α (u, h) = (u, gαβ h).

This fact is important so we record it again:

Let E = (U, gαβ ) be a vector bundle. Then, its frame bundle is

the principal GLn (k)-bundle P = (U, gαβ ).

Thus, we can use the same data to define either vector bundles or principal GLn (k) bundles.
It is in this sense that we mean they are two sides of the same coin. However, we can
construct vector bundles also from principal bundles with other structure groups.

Definition 1.1.15. Given a principal G-bundle P = (U, gαβ ), and a representation ρ : G →

GLn (k), we define the associated vector bundle P ×ρ k n by

E = (U, ρ(gαβ )) (1.25)

We say that a vector bundle E has a (G, ρ)-structure if E = P ×ρ k n for a principal G-bundle
P.

Example 1.1.16. If G ⊂ GLn (k) is a subgroup, it has the fundamental representation

ι : G ,→ GLn (k). We say that E has a G-structure if it has a (G, ι) structure. This means
that we can find U and gαβ such that E = (U, gαβ ), where the gαβ take values in G ⊂ GLn (k).
For example, an orientation of M is the same as an SLn (k)-structure on T M .

Example 1.1.17 (Adjoint bundle). Let G be a Lie group and g be its Lie algebra. Then G
acts on g via the adjoint action (if G is a matrix group then this action is given by conjugation
g · X = gXg −1 ). Hence, for every principal G-bundle P we have the adjoint bundle P ×ρ g.
We denote this bundle by Ad P .

18
1.1.3 Connections on vector bundles

Very roughly, a connection on a fiber bundle - a natural generalization of vector and principal
bundles - is a consistent way to move from one fiber in the bundle to the other. The
connections really lie at the heart of gauge theory in both mathematics and physics. The
concept of connection exists for both vector bundles and principal bundles, but since those
concepts are related (as we explained above), so are the concepts of their connections, as we
will see below. We start with the concept of a connection on vector bundles.

Definition 1.1.18 (Connection on vector bundle). Let π : E → M . Then a connection on

E is a linear map
∇ : Γ(E) → Γ(T ∗ M ⊗ E)

such that, for f ∈ C ∞ (M ) and σ ∈ Γ(E), the Leibniz rule

∇(f σ) = df ⊗ σ + f ∇σ (1.26)

holds.

The connection ∇ induces a covariant derivative along vector fields on sections on E.

Definition 1.1.19 (Covariant derivative). Let ∇ be a connection on the vector bundle E

over M . Let X be a vector field on M . Then, the map ∇X : Γ(E) → Γ(E) given by

∇X σ = ιX ∇σ (1.27)

is called the covariant derivative of σ along X (with respect to ∇).

The Leibniz rule for the covariant derivative is

∇X (f σ) = (LX f )σ + f ∇X σ. (1.28)

Example 1.1.20. On the trivial bundle M × k n we have a connection given by the de Rham
differential (f1 , . . . fn ) 7→ (df1 , . . . dfn ). This connection is called the trivial connection.

Proposition 1.1.21. If it is not empty, the space of connections AE is an affine space

modeled on the vector space Γ(M, T ∗ M ⊗ E ∗ ⊗ E) = Γ(M, T ∗ M ⊗ End E) = Ω1 (End E).

19
Proof. First observe that if ∇1 and ∇2 are connections on E, then their difference A =
∇1 −∇0 satisfies A(f σ) = f A(σ). Hence A defines a vector bundle morphism E → T ∗ M ⊗E.
It follows that A ∈ Γ(Hom(E, T ∗ M ⊗ E)) ∼= Ω1 (End(E)).

It is not hard to show that connections always exist, see for instance [Tau11]. In particular,
every connection on the trivial bundle M × k n is of the form ∇ = d + A, where A ∈
Ω1 (End(k n )). In a basis of k n we write3 A(ej ) = Aij ei , then,

∇(f1 , . . . , fn ) = (df1 , . . . dfn ) + (Ai1 fi , . . . Ain fi ).

Thus one can think of A as a 1-form with values in matrices, or, equivalently, as a matrix
of 1-forms Aij . Both viewpoints are sometimes helpful.

If Ψ : E → F is an isomorphism of vector bundles, and ∇ is a connection on E, then the

map ∇ ˇ = (id ⊗ Ψ) ◦ ∇ ◦ Ψ−1 is a connection on F . It is the unique map that makes

∇
Γ(E) Γ(T ∗ M ⊗ E)
Ψ−1 id⊗Ψ (1.29)
ˇ
∇
Γ(F ) Γ(T ∗ M ⊗ F )

commute. In particular, consider a local trivialization Ψα : E Uα

→ Uα × k n of E. Then a
connection ∇ on E induces a connection on Uα × k n , hence an element Aα ∈ Ω1 (Uα , End k n )
and we have
(∇σ)α = dσα + Aα σα (1.30)

(in the second term there is matrix-vector multiplication). This is called the connection
1-form of ∇ in Uα . If Uβ is another local trivialization, one can ask how Aα and Aβ are
related.

Proposition 1.1.22. Let ∇ be a connection on the vector bundle E and Aα , Aβ be the

connection 1-forms on two local trivializations Uα , Uβ . Then

−1 −1
Aβ = gαβ Aα gαβ − (dgαβ )gαβ . (1.31)
3
We follow the Einstein summation convention that repeated indices are summed over. This does not
apply to indices labeling covers (usually α, β, γ).

20
Proof. Let σ ∈ Γ(E). Then, we know that

σβ = gαβ σα (1.32)

and
(∇σ)β = gαβ (∇σ)α . (1.33)
Expanding (1.33) using (1.30), we obtain

dσβ + Aβ σβ = gαβ (dσα + Aα σα )

On the other hand, using (1.32) we obtain

(∇σ)β = dσβ + Aβ σβ = d(gαβ σα ) + Aβ gαβ σα = dgαβ σα + gαβ dσα + Aβ gαβ σα .

We conclude that
Aβ gαβ σα = gαβ Aα σα − dgαβ σα .
Since this holds for all σ, we see that

Aβ gαβ = gαβ Aα − dgαβ

from where the claim follows.

Hence, we can characterize a connection on a bundle E = (U, gαβ ) as a collection of 1-forms

Aα ∈ Ω1 (Uα , End k n ). Notice that End k n is the Lie algebra of GLn (k). This suggest a
natural generalization of the concept of connections to principal bundles, as discussed in the
next Section.

1.1.4 Connections on principal bundles

We start with some definitions. Suppose G is a Lie group with Lie algebra g that acts on a
manifold P from the right. For fixed p ∈ P , there is a map

µp : G → P
g 7→ pg.

The differential of this map at the identity element e ∈ G is a map

(dµp )e : g ∼
= Te G → Tp P
X 7→ (dµp )e X

21
Definition 1.1.23 (Fundamental vector field). Let the Lie group G act on the manifold P
from the right and let X ∈ g. Then, the fundamental vector field X ] on P is the section of
T P defined by
Xp] = (dµp )e X. (1.34)

In particular, G acts on itself by right multiplication. For fixed g ∈ G, this action reads

µg : G → G
h 7→ gh

Hence we have µg = Lg (left multiplication by g). Let X ∈ X. The fundamental vector field
of the right action of G on itself is given by X 7→ (dLg )e X.

Definition 1.1.24 (Maurer-Cartan Form). The Maurer-Cartan Form φ ∈ Ω1 (G, g) is defined

by φg (Xg] ) ≡ X ∈ g, where X ] is the fundamental vector field of the right action of G on
itself.

Remark 1.1.25. From the discussion above it follows that

φg = (dLg−1 )g : Tg G → Te G ∼
= g. (1.35)

In particular, for matrix groups it is given by φg = g −1 dg.

We can now define a connection on a principal bundle.

Definition 1.1.26 (Connection on a principal bundle). Let π : P → M be a principal G-

bundle, and let g be the Lie algebra of G. A connection on P is a 1-form Ω ∈ Ω1 (P, g)
satisfying4

i) For all g ∈ G,
Rg∗ Ω = Adg−1 Ω (1.36)

ii) For all X ∈ g,

Ω(X ] ) = X ∈ g (1.37)
4
Here, Adg : g → g is given by differentiating the map Adg : G → G at the identity. For matrix groups
G ⊂ GLn (k) we have g ⊂ gln (k) and the adjoint action is Adg X = gXg −1 .

22
Notice that here, the 1-form Ω is defined on the total space P . A local section σ : U → P U
defines a local trivialization ψσ : π −1 (U ) → U × G of P by letting ψσ (σ(u)) = (u, 1) and
asking ψσ to be G-equivariant. Explicitly, the map is defined by

ψσ (p) = (π(p), φσ (p)) (1.38)

where φσ (p) is the unique element of G satisfying σ(π(p))φσ (p) = p.

Let σ 0 be another section over U . Then there exists a map g : U → G such that σ 0 (x) =
σ(x)g(x). The following lemma describes the behavior of a connection under such a change
of trivialization.

Lemma 1.1.27. If σ, σ 0 are as above, then

(σ 0 )∗ Ω = Adg−1 σ ∗ Ω + g ∗ φ, (1.39)

where φ ∈ Ω1 (G, g) is the Maurer-Cartan Form introduced above.

Proof. Let x ∈ U , and v ∈ Tx M . Then ((σ 0 )∗ Ω)x v = Ωσ0 (x) dσx0 v. On the other hand we can
write σ 0 as the composition
(σ,g) µ
U P ×G P

x (σ(x), g(x)) σ(x)g(x)

We first compute the pullback µ∗ Ω to P × G. For this, note that

(dµp )g w = (dµpg )e (dLg−1 )w

(this follows from the chain rule). Then, we have

(µ∗ Ω)(p,g) (v, w) = Ωpg (dµp )g w + Ωpg (dRg )p v

= Ωpg (dµpg )e (dLg−1 )g w +(Rg∗ Ω)p v
| {z }
((dLg−1 )g w)]

= (dLg−1 )g w + Adg−1 Ωp v = φg w + Adg−1 v

where in the last equality we used the two properties of a connection. Now, the first term
is exactly φg . Pulling back to U with (s, g), we obtain the result since pulling back with σ
commutes with the adjoint action of g, which acts only on the Lie algebra factor of Ω.

23
The next proposition establishes the relationship of this definition with the one of a connec-
tion on a vector bundle.

Proposition 1.1.28. Let P = (U, gαβ ) be a principal bundle with structure group G ⊂
GLn (k), i.e. G is a matrix group. Then a connection on P is equivalent to a collection of
1-forms Aα ∈ Ω1 (Uα , g) such that
−1 −1
Aβ = gαβ Aα gαβ − dgαβ gαβ . (1.40)

Proof. Suppose we are given a connection Ω on P and let Uα ∈ U. Consider the constant
section σα : Uα → Uα × G, u 7→ (u, 1). Then, we define Aα := (Ψ−1 ∗
α ◦ σα ) Ω. Now, notice that
the section σα over the Uβ is given by gαβ . Hence
−1
σβ = σα gαβ .
−1
We can now apply Lemma 1.1.27 for gαβ . Notice also that we have for any map g : U → G
that 0 = d(gg −1 ) = dgg −1 + gd(g −1 ) and hence d(g −1 ) = −g −1 dgg −1 . This implies that for
a matrix group, we have φg−1 = gd(g −1 ) = −dgg −1 . This proves that a connection Ω is
described by such 1-forms in a local trivialization.
Conversely, assume that we are given a family of such 1-forms. Then we set Ωα (u, g) :=
Adg−1 Aα + φg on Uα × G and define Ω on π −1 (Uα ) as Ψ∗α Ωα . We then glue together the
connection using a partition of unity. The resulting 1-form Ω is a connection since the local
pieces are, and the conditions (1.36) and (1.37) are convex.

We have the following corollary:

Corollary 1.1.29. A connection ∇ on a vector bundle E induces a connection Ω on the

bundle of frames Fr(E) and vice versa.

Hence, one can study connections on vector bundles by studying connections on principal
bundles. This will be our approach in this course.

Curvature

An important notion associated to a connection is the concept of curvature. For this, we

need the concept of Lie Bracket on Lie algebra-valued forms, which is defined on elements

24
of the form A = α ⊗ ξ, B = β ⊗ ξ 0 , where α ∈ Ωk (M ), β ∈ Ωl (M ), ξ, ξ 0 ∈ g, by

[·, ·] : Ωk (M, g) × Ωl (M, g) → Ωk+l (M, g)

[α ⊗ ξ, β ⊗ ξ 0 ] := α ∧ β ⊗ [ξ, ξ 0 ] (1.41)

and extended bilinearly. For simplicity, we now assume that g is a matrix Lie algebra, but
all constructions mentioned here work also for general Lie algebras. we have [ξ, ξ 0 ] = ξξ 0 −ξ 0 ξ
and then
[A, B] = A ∧ B − (−1)|A||B| B ∧ A = −(−1)|A||B| [B, A] (1.42)

where the wedge product operation is defined by matrix multiplication: If A = α ⊗ ξ, B =

β ⊗ ξ 0 as above, then
A ∧ B = α ∧ β ⊗ ξξ 0 (1.43)

The Lie bracket satisfies

d[A, B] = [dA, B] + (−1)|A| [A, dB] (1.44)

[A, [B, C]] = [[A, B], C] + (−1)|A||B| [B, [A, C]] (1.45)

Now, the curvature is easily defined from the abstract viewpoint on connections:

Definition 1.1.30 (Curvature). Let Ω ∈ Ω1 (P, g) be a connection on a principal G-bundle

π : P → M . Then, the curvature of Ω is the 2-form F ∈ Ω2 (P, g) defined by
1
FΩ = dΩ + [Ω, Ω] (1.46)
2

An important condition on connections is flatness:

Definition 1.1.31 (Flat connection). A connection Ω ∈ Ω1 (P, g) is called flat if its curvature
vanishes: FΩ = 0.

We summarize some properties of the curvature as exercises.

Exercise 4. Let U be a trivializing cover of P . Let (Uα , ψα ) be a local trivialization. Then
there is an associated local section σα : Uα → π −1 (Uα ) given by σα (x) = ψα−1 (x, 1). Denote
Fα := (σα )∗ FΩ = dAα + 21 [Aα , Aα ]. Show that

−1
Fβ = gαβ Fα gαβ (1.47)

25
It follows that the Fα define a section F ∈ Ω2 (M, Ad P ).
Exercise 5. Let E → M be a vector bundle and let ∇ be a connection on E. Define the
two-form F ∇ ∈ Ω2 (M, End E) by

F ∇ (X, Y ) = ∇X ∇Y − ∇Y ∇X − ∇[X,Y ] . (1.48)

Show that this is the curvature 2-form of the associated connection on Fr(E).
Hint: Work over a trivializing chart and remember the formula for the de Rham differential
of a 1-form:
dω(X, Y ) = Xω(Y ) − Y ω(X) − ω([X, Y ]).

Exterior Derivative

A connection on a principal bundle P → M induces an exterior derivative on Ad P -valued

differential forms. In a trivializing chart Uα , it is defined by

(dΩ ω)α = dωα + [Aα , ωα ] (1.49)

Proposition 1.1.32. i) The exterior derivative in local trivializations by (1.49) defines a

map

dΩ : Ωk (M, Ad P ) → Ωk+1 (M, Ad P )

ω 7→ dΩ ω

ii) We have
dΩ dΩ ω = [FΩ , ω] (1.50)

iii) The curvature satisfies

dΩ FΩ = 0, (1.51)

the Bianchi identity.

Proof. i) One simply checks by direct computation that

−1
(dΩ ω)β = gαβ (dΩ ω)α gαβ .

26
 ii) By the first point, it is enough to check this in a trivializing chart. Here, again the proof
is a simple computation:

(dΩ dΩ ω)α = dΩ (dωα + [Aα , ωα ])

= d(dωα ) + [Aα , dωα ] + d[Aα , ωα ] + [Aα , [Aα , ωα ]]
1
= [Aα , dωα ] + [dAα , ωα ] − [Aα , dωα ] + [[Aα , Aα ], ωα ]
2
= [Fα , ωα ]

where we have used (1.44) and (1.45).

iii) Again one can check this in a trivializing chart. Here we simply compute

(dΩ F )α = dFα + [Aα , Fα ]

1 1
= d(dAα ) + d[Aα , Aα ] + [Aα , dAα ] + [Aα , [Aα , Aα ]]
2 2
The last term vanishes due to (1.45) and the other terms cancel due to (1.44).

In particular, if Ω is flat, we have that dΩ : Ω• (M, Ad P ) → Ω• (M, Ad P ) squares to zero and

we can define the cohomology

H • (M, Ad P ) := ker(dΩ )/im(dΩ ). (1.52)

1.2 Chern-Simons action functional

After establishing the necessary preliminaries, let us turn to the definition of the Chern-
Simons action functional. The original reference is [CS74]. These notes closely follow the
review [Fre95].

1.2.1 The Chern-Simons 3-form

We first fix a compact, connected and simply connected matrix group G ⊂ GL(n) (the prime
example being SU (n), n ≥ 2)) with Lie algebra g. Next, fix an ad-invariant non-degenerate

27
symmetric bilinear form h·, ·i on g (the prime example being the Killing form on SU (n),
which is a multiple of the trace). Here ad-invariant means that

hadx y, zi = −hy, adx zi (1.53)

(equivalently, h·, ·i is invariant under the adjoint action of G on g). It follows that

h[·, ·], ·i : ∧3 g → R

is a Lie-algebra 3-cocycle (i.e. it is completely antisymmetric in all 3 arguments and closed

under the Chevalley-Eilenberg differential).

Definition 1.2.1. Let G ,→ P  M be a principal G-bundle and Ω a connection on P .

Then we we define the first Pontryagin form of P to be
1
p1 (Ω) = hFΩ , FΩ i ∈ Ω4 (P ). (1.54)
2
Proposition 1.2.2. For any trivializing cover U = {Uα }, the pullbacks σα∗ p1 (Ω) piece to-
gether into a global 4-form pM 4
1 (Ω) ∈ Ω (M ) which is closed.

Proof. If one changes the trivializing chart the pullback of the curvature gets conjugated,
hence by ad-invariance we have pM M
1 (Ω)α = p1 (Ω)β . Hence the pullbacks piece together into
a global 4-form. The closedness follows from the Bianchi identity and ad-invariance.

Definition 1.2.3. The cohomology class [pM 4

1 (Ω)] ∈ H (M ) is called first Pontryagin class
of P .

On P , the form p1 (Ω) is not only closed, but also exact:

Definition 1.2.4. The Chern-Simons 3-form cs(Ω) ∈ Ω3 (P ) is defined by

1 1
cs(Ω) := hΩ, dΩi + hΩ, [Ω, Ω]i (1.55)
2 6
1 1
= hΩ, FΩ i − hΩ, [Ω, Ω]i. (1.56)
2 12

This Chern-Simons form is a primitive of the Pontryagin form:

Proposition 1.2.5. We have

dcs(Ω) = p1 (Ω). (1.57)

28
Proof. Exercise (use the Bianchi identity!).

It is interesting to observe how the Chern-Simons form transforms under gauge transfor-
mations. If ϕ : P → P is a gauge transformation, i.e. an automorphism of P , there is an
associated map gϕ : P → G defined by the requirement

ϕ(p) = p · gϕ (p). (1.58)

Proposition 1.2.6. We have

1 1
ϕ∗ cs(Ω) = cs(Ω) + d(gϕ Ωgϕ−1 ∧ gϕ−1 dgϕ ) − gϕ∗ (hφ, [φ, φ]i), (1.59)
2 2
where φ ∈ Ω1 (G, g) is the Maurer-Cartan form defined in 1.1.24.

Proof. The proof is a straightforward computation and left to the reader as an exercise.

This transformation behavior is very different from the one of the Pontryagin 4-form. In
particular, usually there is no globally defined Chern-Simons 3-form on M . However, we can
ask ourselves if the cohomology class of the Pontryagin 4-form [pM 4
1 (Ω)] ∈ H (M ) (called the
Pontryagin class) is trivial. A particular case when this happens is when the bundle admits
a global section (i.e. it is trivial). In that case we conclude that the class of the Pontryagin
4-form is trivial in cohomology. It follows that if the cohomology class of the Pontryagin
4-form is non-trivial, then the bundle cannot be trivial - this is an example of the use of
characteristic classes.

1.2.2 The action functional

The following is a relatively simple but crucial fact in low-dimensional gauge theory.

Lemma 1.2.7. Let G be a connected and simply connected Lie group. Then any principal
G-bundle over a manifold M of dimension less than or equal to 3 is trivializable.

Proof. We give a proof for the interested reader using a little algebraic topology and obstruc-
tion theory. Every principal G-bundle on a manifold M corresponds to a map M → BG,

29
where BG denotes the classifying space of G.5 Homotopic maps correspond to isomorphic
bundles. But the fact that π0 (G) = π1 (G) = π2 (G) = 0 (vanishing of first two homotopy
groups follows from assumptions on G, vanishing of π2 (G) is a general fact for connected Lie
groups6 ) implies that π1 (BG) = π2 (BG) = π3 (BG) = 0. Since BG is connected, we always
have π0 (BG) = 0 and hence7 for any manifold of dimension less than or equal to 3 we have

[M → BG] ∼
= ∗,

i.e. all maps are homotopic to the constant map, which corresponds to the trivial bundle.
Hence all principal bundles are isomorphic to the trivial one.

This fact is very much not true if the Lie group is not simply connected (e.g. G = U (1)). In
that case, Chern-Simons theory becomes a lot more complicated. See [Fre02] for a discussion.

Definition 1.2.8. Let M be a compact oriented 3-manifold and G ,→ P  M . Let

s : M → P be a global section of P (guaranteed to exist by the previous Lemma). Then we
define the Chern-Simons action functional by
Z Z
∗ 1 1
SCS [s, Ω] = s cs(Ω) = hA, dAi + hA, [A, A]i (1.60)
M M 2 6
where we defined A := s∗ Ω.

As a consequence of Proposition 1.2.6, this action functional is almost invariant under gauge
transformations (automorphisms of P ).

Proposition 1.2.9. Let M be a closed manifold and ϕ : P → P be a gauge transformation.

Let s be a global section of P and define g = gϕ ◦ s. Then we have
Z
∗ 1
SCS [ϕ ◦ s, Ω] = SCS [s, ϕ Ω] = SCS [s, Ω] − g ∗ hφ, [φ, φ]i (1.61)
12 M

Proof. This follows directly from Equation (1.59).

5
The classifying space BG is the base of the “universal bundle” EG → BG which is a principal bundle
with the property that EG is contractible.
6
See e.g. [BD13, Chapter V].
7
All the obstructions for two maps not to be homotopic vanish. For a deeper discussion of obstruction
theory see [Hus93].

30
Note that if s and s0 are any two section of P , then there exists a global gauge transformation
ϕ with ϕ ◦ s = s0 : Concretely, ϕ = Φs0 ◦ Φ−1
s where Φs : M × G → P is the trivialization
given by Φs (x, g) = s(x)g. Hence (1.61) tells us two things: How to relate the action
functionals in different trivializations, and how the action functional transforms under gauge
transformations.
Proposition 1.2.9 motivates the following assumption:
Assumption 1.2.10. The bilinear form h·, ·i is such that
1
hφ, [φ, φ]i ∈ H 3 (G, Z) ⊂ H 3 (G, R) (∼
= H 4 (BG, Z)) (1.62)
12

This assumption implies in particular that

Z
g ∗ hφ, [φ, φ]i ∈ 12Z (1.63)
M

and hence that for every integer k the exponentiated Chern-Simons action

e2πikSCS [s,Ω] ≡ e2πikSCS [Ω]

is independent of the choice of trivialization and invariant under gauge transformations.

1.3 Critical Points

Now that we know the space of fields of the theory (the space of connections on the (unique
up to isomorphism) principal G-bundle on M ) and the action functional (the integral of the
Chern-Simons 3-form) the next step to understand the theory is to understand the critical
points of the action functional. We do not enter into the technical details of derivatives in
infinite dimensions here, but rather just define a critical point to be a connection A such
that for all B ∈ Ω1 (M, Ad P )8

d
SCS [A + tB] = 0. (1.64)
dt t=0
8
The space of connections is an affine space with espace vectoriel directeur Ω1 (M, Ad P ), hence a tangent
vector to a connection A is an element B ∈ Ω1 (M, Ad P ), and a curve with this tangent vector at 0 is simply
A + tB.

31
A quick computation shows that
Z
1
SCS [A + tB] = SCS [A] + t hB, dA + [A, A]i + O(t2 ), (1.65)
M 2
whence we conclude that
Z Z
d 1
SCS [A + tB] = hB, dA + [A, A]i = hB, FA i. (1.66)
dt t=0 M 2 M

This is often rewritten as Z

δS[A] = hδA, FA i. (1.67)
M
It follows that the critical points of the action functional are precisely the flat connections.
Since the curvature FA is a two-form with values in the adjoint bundle (Exercise 4), flat
connections are sent to flat connections under gauge transformations and it makes sense to
ask about the quotient of the space of flat connections under gauge transformations.9

Definition 1.3.1. For a manifold M and a Lie group G, we define 10 the moduli space of flat
G-connections by
{(P, Ω)|P principal G-bundle, Ω flat connection on P }
M F C(M, G) = (1.68)
{isom. of principal G-bundles}

This is space of considerable interest in topology and geometry, but in this guise completely
inaccessible. We will give another characterization in the next section.

1.4 The representation variety

The moduli space of flat connections is a complicated and intriguing object. In this section
we sketch a proof of the often used fact that it has an equivalent characterization that shows
how it is determined by the topology of M and the algebra of the group G:

Theorem 1.4.1. There is a bijection

Hom(π1 (M ), G)
M F C(M, G) ∼
= (1.69)
G
where the group G acts on group homomorphisms with codomain G by conjugation.
9
To be precise in what follows, here we relax the notion of gauge transformation to the one of bundle
isomorphism (instead of automorphism).
10
We refer to [FQ93] for a deeper discussion on the technicalities of this quotient.

32
This result is nothing less than astonishing: On the left-hand we are identifying solutions to
complicated non-linear differential equation via the action of an infinite-dimensional group,11
while the right-hand side depends on M only through the first fundamental group - a rather
crude topological invariant of M ! Of course, the quotient on the right-hand side often turns
out to be quite complicated too, but the theorem certainly provides a vast improvement in
understanding M F C(M, G). The rest of this section is devoted to a sketch of the proof of
Theorem 1.4.1, which is highly instructive in itself, mainly following [Tau11].

1.4.1 The horizontal distribution of a connection

An important tool to study a connection is its horizontal distribution.

Definition 1.4.2. Let Ω ∈ Ω1 (P, g) be a connection on P , then its horizontal distribution

is the subbundle of T P given by

HΩ = ker Ω ⊂ T P (1.70)

Remark 1.4.3. The dimension of this kernel at every point is dim M , hence this is indeed a
subbundle.
12
Proposition 1.4.4. 1. Denote the kernel of dπ : T P → T M by V P . Then we have

T P = V P ⊕ HΩ . (1.71)

2. The horizontal distribution is equivariant with respect to the right G-action,

(HΩ )pg = (HΩ )p · g (1.72)

(here the action on the right hand side is the derivative of the right G-action on P ).

Proof. 1. Let p ∈ P , the map ι : g → Vp P , X 7→ (X ] )p (the fundamental vector field of

X evaluated at p) is an isomorphism (it is injective because the G-action is free, and
they have the same dimension). Since Ωp (X ] ) = X, we have that ι ◦ Ωp is a projection
to Vp P ⊂ Tp P , and ker(ι ◦ Ωp ) = ker(Ωp ) = (HΩ )p is a complement of Vp P .
11
Groupoid, to be completely precise.
12
This is called the vertical tangent bundle or tangent bundle along the fibers of P .

33
 2. For v ∈ Tp P , we have (dRg )p v ∈ Tpg P and

Ωpg ((dRg )p v) = (Rg∗ Ω)p v = Adg−1 (Ωp (v)). (1.73)

Since (dRg )p , Adg−1 are linear isomorphisms, we conclude that v ∈ ker Ωp ⇔ (dRg )p v ∈
ker Ωpg . This proves the claim.

Remark 1.4.5. On any smooth fiber bundle P  M , one can define a connection as a sub-
bundle H of T P satisfying (1.71) (and define the connection one-form by the corresponding
projection). Over principal bundles one additionally requires (1.72).13 . This definition is
equivalent to the Definition 1.1.26. Working out the details is an instructive exercise.

Flatness of connections corresponds to an important property of the distribution:

Proposition 1.4.6. The horizontal distribution is integrable if and only if Ω is flat.

Proof. Let X, Y ∈ Γ(HΩ ) (i.e. X, Y are vector fields tangent to HΩ ). By definition,

this means that Ω(X) = Ω(Y ) ≡ 0. An easy computation then shows that Ω([X, Y ]) =
−FΩ (X, Y ), which proves the statement.

1.4.2 Parallel transport

Any connection defines a notion of parallel transport as follows. Again let P be a principal
G-bundle over M .

Definition 1.4.7. Let p ∈ P and γ : [0, 1] → M and path with p ∈ Pγ(0) . Then a curve γ̃(t)
˙
is called a horizontal lift of γ if π(γ̃(t)) = γ(t) and γ̃(t) ∈ (HΩ )γ̃(t) for all t ∈ [0, 1].

It follows from the basic theory of differential equations that horizontal lifts always exist and
are unique (see e.g. [KN96]).
13
Some sources do not ask for this, and call connections satisfying (1.72) principal connections, but since
we only care for this type of connections, we do not need to make this distinction.

34
Definition 1.4.8. Let p ∈ P and γ : [0, 1] → M with p ∈ Pγ(0) . Then we define the parallel
transport along γ by
P t(Ω, γ)(p) = γ̃(1) (1.74)

The parallel transport along γ defines a map

P t(Ω, γ) : Pγ(0) → Pγ(1) .

A crucial result is the following.

Theorem 1.4.9. If Ω is flat, the parallel transport along a path γ depends only on its
homotopy class.

Instead of giving a detailed proof, for which we refer to [KN96] or [Tau11], we give an easy
example that gives an idea of the corresponding phenomenon.

Example 1.4.10. Take G = R and consider the trivial G-bundle P = M × R → M . Any

connection on P is of the form Ω = dx − A, where x is the coordinate on R and A ∈ Ω1 (M ).
If γ : [0, 1] → M is a path we claim that the horizontal lift starting at p = (γ(0), x0 ) is
˙ ∂
R
γ̃(t) = (γ(t), x0 + γ|[0,t] A). Indeed, we have γ̃(t) = (γ̇(t), Aγ(t) (γ̇(t)) ∂x ). It follows that
˙
Ω(γ̃(t)) = −A(γ̇(t)) + A(γ̇(t)) = 0. Since the horizontal lift is unique, we conclude that
parallel transport along γ is given by
Z
(γ(0), x0 ) 7→ (γ(1), x0 + A).
γ

Proving that the parallel transport depends only on the homotopy class is equivalent to
proving that the parallel transport around contractible loops is trivial. If γ is a contractible
R R R
loop then γ bounds a disk D ⊂ M and we have γ A = D dA = D FA , since for abelian
groups we have FA = dA. T his shows that the parallel transport along contractible loops is
trivial if and only if FA = 0. The same basic idea also applies in the non-abelian case, but
some more involved concepts are needed.

Definition 1.4.11. Let γ be a closed loop and p ∈ Pγ(0) . Then we define the holonomy of
Holp (Ω, γ) ∈ G by
P t(Ω, γ)(p) = p · Holp (Ω, γ). (1.75)

Lemma 1.4.12. If q = p · g, then Holq (Ω, γ) = g −1 Holp (Ω, γ)g.

35
Proof. It is a consequence of equivariance of the horizontal distribution (1.72) that the
parallel transport commutes with the right G-action. Therefore

P t(Ω, γ)(q) = P t(Ω, γ)(p · g) = P t(Ω, γ)(p) · g

= p · Holp (Ω, γ) · g = p · (g · g −1 ) · Holp (Ω, γ) · g = q · g −1 Holp (Ω, γ)g.

1.4.3 The isomorphism

Fix a point p ∈ P and let x = π(p) ∈ M . Let Ω be a flat connection on P . Then, one can
define a representation

ρP,Ω,p : π1 (X, x) → G
[γ] 7→ Holp (Ω, γ) (1.76)

Lemma 1.4.13. The conjugacy class of ρP,Ω,p does not depend on p (in particular, not on
x).

Proof. If we change p in the fiber over x this follows from Lemma 1.4.12. Dependence on
x is slightly more subtle since in principle also the fundamental group changes (up to inner
isomorphism). We refer to [Tau11] for a proof.

ρP,Ω,p is called the holonomy representation of Ω at p.14 We then have the following theorem,
which is the more precise version of Theorem 1.4.1. Denote

Af lat = {(P, Ω)|Ω flat connection on P }.

Theorem 1.4.14. The map

T : Af lat → Hom(π1 (X, x), G)/G

(P, Ω) 7→ [ρP,Ω,p ]G
14
Sometimes dependence on the point p is dropped and one then understands the quotient by conjugation.
If Ω is not flat one can still define the holonomy map, but it does not descend to π1 : One then usually speaks
of the image of the holonomy map at p, which is called the holonomy group of Ω at p (the same remark
about the basepoint applies).

36
is independent of x and p, and the gauge equivalence class of (P, Ω). It descends to an
isomorphism on the quotient:

T : M F C(M, G) −→ Hom(π1 (X, x), G)/G (1.77)

The proof, for which we again refer to the literature, depends on the construction of an
inverse map. This is given by assigning to a representation ρ : π1 (X) → G the principal
G-bundle P defined by P = X̂ ×ρ G = X̂ × G/ ∼. Here X̂ is the universal cover of X on
which π1 (X) acts by deck transformations, and (x̂, g) ∼ (π · x̂, ρ(π)−1 g).

1.4.4 Topology and smooth structure

To put a topology on M F C(M, G) we use that the fundamental group π1 (M ) is finitely gener-
ated if M is compact. If x1 , . . . , xn are the generators of π1 (M ), then any map ρ : π1 (M ) → G
is completely determined by the images (ρ(x1 ), . . . , ρ(xn )) ∈ Gn . Thus we can identify
Hom(π(M ), G) with a subset of Gn , and equip it with the subspace topology. Consequently,
one can endow M F C(M, G) ∼ = Hom(π1 (M ), G)/G with the quotient topology.15 One can
check that this topology is independent of the choice of generators.
Certain points in the moduli space of flat connections have neighborhoods that admit a
smooth structure, where “smoothness” means the following: A map

f : U ⊂ Hom(π1 (X), G) → R

is smooth if there is a smooth map f˜: V → R such that U ⊂ V ⊂ Gn and f = f˜|U .

To this point, we recall that a flat connection Ω gives a differential dΩ on Ω• (M, Ad P )
given in a trivializing chart by ω 7→ dω + [A, ω]. A map f : U → R on an open subset
U ⊂ M F C(M, G) is smooth if and only if f ◦ π : π −1 (U ) → R is smooth. Finally, a
point x ∈ M F C(M, G) is smooth if and only if there is a neighborhood U of x and a
homeomorphism φ : U → φ(U ) ⊂ RN such that f : U → R is smooth if and only if f ◦ φ−1
is smooth. We refer to [Wal92] for a proof of the following.
15
One could also endow it with the quotient topology of the defining quotient. It has been shown that
these topologies agree (i.e. the map T is a homeomorphism if one endows the domain with the quotient
topology and the codomain with the topology discussed in this section).

37
Theorem 1.4.15. 1. A point [Ω] ∈ M F C(M, G) is smooth if and only if HΩ0 (M, Ad P ) =
H 0 (M ) ⊗ Z(g), where Z(g) is the center of g.

2. The Zariski tangent space to M F C(M, G) at Ω is HΩ1 (M, Ad P ). In particular, at

smooth points, [Ω] ∈ M F C(M, G) has a neighborhood diffeomorphic to an open subset
of HΩ1 (M, Ad P ).

38
Section 2

Perturbative Quantization

After considering the classical Chern-Simons theory, in this Section we explain the methods
of perturbative quantization that we will later use for Chern-Simons theory. The main idea
R i
is to first understand the behavior of a finite-dimensional integral I(~) = X e ~ S µ as ~ → 0.
We will see that it it is completely determined by the behavior of S at its critical points.
We will then arrive at a formula describing the asymptotic behavior of I(~) as ~ → 0. The
idea is then to use this formula as the definition of the “asymptotic behavior” of the path
integral, where the integration domain is infinite-dimensional. As we shall see, this is not
always a straightforward procedure.
We will see that even in the finite-dimensional model, extra care is needed when the critical
points of the action functional are not isolated. In particular, this happens when there is a
symmetry acting on the integration domain leaving the space of fields invariant (in this case
there is an entire orbit consisting of critical points). We will discuss the remedies to this
problem that are usually used in physics: The Faddeev-Popov ghosts and the BRST and BV
formalisms.
A couple of references explaining the perturbative approach, discussing Chern-Simons theory
or gauge theory in particular, are [Pol05],[Res10],[Mne17]. In this Section we follow closely
the book by Mnev [Mne17], in particular the Sections 3 (Feynman diagrams) and 4 (Batalin-
Vilkovisky formalism) there (we will sometimes point to particular sections for more details
on certain points). We start with the asymptotics of oscillatory integrals, which is a classic
topic of microlocal analysis.

39
2.1 Asymptotics of oscillatory integrals

We recall briefly a few facts about derivatives on manifolds. If X is a manifold and S : X → R

is a smooth function, then a point p ∈ X is called critical if the differential dS : Tp X → R
vanishes. In that case, there is a coordinate-independent bilinear form Hp S called the Hessian
of S at p. With respect to any coordinate system it takes the form
∂ 2S
(Hp S)ij = (2.1)
∂xi ∂y j p

Under a change of coordinates it transforms as a bilinear form on Tp X. In particular,

its determinant is not well-defined but depends upon choice of coordinates.1 The higher
derivatives, however, do not transform tensorially, but rather as jets.2 We say that p is a
non-degenerate critical point of S if the Hessian is non-degenerate.
Also, let us briefly explain the notation for asymptotic behavior: Namely, for f a continuous
function on (0, a) for some a > 0 we define
f (~)
f (~) '~→0 O(~N ) : ⇔ ∃CN ∈ R, ε > 0 : ≤ CN ∀0 < ~ < ε (2.2)
~N
Similarly one defines

f (~) '~→0 g(~) + O(~N ) : ⇔ f (~) − g(~) '~→0 O(~N ). (2.3)

2.1.1 Stationary phase formula

The first result is the following theorem, also known as the stationary phase formula.

Theorem 2.1.1. Let X be a compact manifold with dim X = n, µ ∈ Ωn (M ) a volume form

and S : X → R have finitely many non-degenerate critical points. Then we have
iπ
e 4 sign Hx0 S
Z
i n i
X
S S(x ) 1+n/2
I(~) := e ~ µ '~→0 (2π~) 2 e~ 0
1 µx0 + O(~ ) (2.4)
X x0 ∈Crit(S)
| det H x 0 S| 2

where one chooses coordinates y 1 , . . . , y n around x0 to define the determinant of the Hessian
and µx0 is given by µ(x0 ) = µx0 dy 1 ∧ . . . ∧ dy n .
1
Under coordinate changes, its determinant changes with the square of the determinant of the coordinate
change, as opposed to the determinant of an endomorphism, which is invariant under change of coordinates.
2
In the usual tautological way, jets are defined to be section of bundles that transform like derivatives of
a function under changes of coordinates.

40
Note that the expression on the right hand side is independent of the choice of coordinates:
The signature sign Hx0 is independent under coordinate change, and the changes in det Hx0
and µx0 cancel.
The proof of this theorem follows from a sequence of lemmata.

Lemma 2.1.2 (Fresnel integrals). Let Q be a symmetric bilinear form on Rn , and Q0 any
positive definite symmetric bilinear form. Then
iπ
e 4 sign Q
Z
n iQ(x,x)−εQ0 (x,x) n/2
lim d xe =π (2.5)
ε→0+ Rn | det Q|1/2

The proof of this lemma follows from the one-dimensional case, which is proven by standard
complex analysis, together with a change of coordinates. The details are left to the interested
R
reader as an exercise. We interpret the left-hand side as the definition of Rn eiQ(x,x) (which
is only conditionally convergent).
i
Lemma 2.1.3. Let g ∈ Cc∞ (R) and consider I(~) =
R
R
g(x)e ~ x dx. Then

I(~) '~→0 O(~∞ ) ⇔ I(~) '~→0 O(~N ) ∀N ∈ N. (2.6)

The proof is just integration by parts:

Proof. Let N ∈ N, then

Z  N Z
I(~) d i
x
= g(x) −i e ~ dx ≤ g (N ) (x) dx = : CN .
~N R dx R

i
Here we are using the “phase function” (the multiplier of ~
in the exponent) f (x) = x,
which has no critical points on R. The conclusion is that the integral vanishes faster than
any power of ~. This is true also in the multi-dimensional case:

Lemma 2.1.4. Let g ∈ Cc∞ (Rn ) and f ∈ C ∞ (Rn ) such that f has no critical points on the
support of g. Then Z
i
dn xg(x)e ~ f (x) '~→0 O(~∞ ) (2.7)
Rn

41
Proof. W.l.o.g. we can assume that f has no critical points at all. Thus the sets f −1 (y), y ∈ R
are embedded submanifolds in Rn and we can rewrite the integral as
Z Z Z
i i
n f (x) y
d xg(x)e ~ = dye ~ g(x)dvolf −1 (y) (x).
R f −1 (y)

Denoting the integral over the fiber by h(y), we see that h is a smooth function and supp h ⊂
f (supp g) which is compact as the image of a compact set. Hence we can apply the previous
lemma to conclude the proof.

Finally, let us look at the situation where there is a single non-degenerate critical point. This
can be described by the following situation:

Lemma 2.1.5. Let Q be a non-degenerate quadratic form on Rn and g ∈ S(Rn ). Let gN be

the N -th Taylor expansion of g. Define
Z
i
I(~) := (g(x) − gN (x))e ~ Q(x,x) .
Rn

Then
N +2
c
I(~) '~→0 O(~n/2+b 2 ) (2.8)

Proof. The proof goes again through integration by parts. Let Q(x, x) = 12 Qij xi xj , then
 
∂ i Q(x,x) 2i i i
Q(x,x)
e ~ = x Qij e ~ .
∂xj ~

We conclude that
−i −1 ij 1 ∂
D= (Q ) i j
2 x ∂x
satisfies
i 1 i Q(x,x)
De ~ Q(x,x) = e~
~
and hence Z
I(~) i
= dn x(g(x) − gN (x))Dm e ~ Q(x,x) .
~m Rn
R R
To integrate by parts we have to find the operator DT such that Rn
f Dgdn x = Rn
(DT f )gdn x,
this operator is given by
i ∂ 1
DT = (Q−1 )ij j j
2 ∂x x

42
as one immediately verifies. Since gN is the N -th Taylor approximation of g, we have
|g(x) − gN (x)| = C|x|N +1 (1 + O(|x|)), where C is a constant (possibly 0). Applying the
operator DT decreases the power of the absolute value by 2 (once by dividing and once by
taking a derivative). Thus

|(Dm )T (g − gN )(x)| = C 0 |x|N +1−2m (1 + O(|x|)).

It is an elementary exercise (using e.g. polar coordinates) to verify that, if k ∈ Z, ε > 0 and
n ∈ N, we have that Z
|x|k dn x < ∞ ⇔ k > −n.
Dε (0)

I(~) N +n+1
We conclude that ~m
≤ Cm if N + 1 − 2m > −n ⇔ m < 2
. This is almost the
estimate claimed in the Lemma, for the remaining part (which is not too difficult) we refer
to [Mne17, Chapter 3].

We are now ready to prove the stationary phase formula (2.4).

Proof of Theorem 2.1.1. To prove the stationary phase formula, cover the manifold X with
charts {Uα } such that each of the critical points x1 , . . . , xN is contained in exactly one chart.
Let {ψα } be a partition of unity subordinate to the open cover {Uα }. Then, we have
Z XZ
i i
S
e µ=
~ dn y µα (y)ψα (y)e ~ S(y)
X α Uα
XZ i
(compact support of ψα ) = dn y µα (y)ψα (y)e ~ S(y)
α Rn
N Z   
X i 1
Lemma 2.1.4 ' n
d y µα (y)ψα (y) exp i j
S(xk ) + (Hxk )ij y y + P (y) + O(~∞ )
k=1 Rn ~ 2

Here we have denoted P (y) the terms of degree at least 3 in the Taylor series of S. Applying
i
Lemma 2.1.5 with g = µψα e ~ P (y) and N = 0 we find
N Z N Z
i i i
(Hxk S)ij y i y j i j
X X
n S(xk )
d y ge 2~ '~→0 e ~ dn y µ(x0 )e 2~ (Hxk S)ij y y + O(~1+n/2 )
k=1 Rn k=1 Rn
N iπ
X i
S(xk ) n/2 e 4 sign Hxk S
Lemma 2.1.2 = e ~ (2π~) µ(x0 ) + O(~1+n/2 )
k=1
| det Hxk S|1/2

which concludes the proof.

43
2.1.2 Higher order corrections

With just a little extra effort, we can derive a closed formula for the asymptotic behavior
to all orders in ~. The importance of Fresnel moments becomes obvious from the following
Lemma.

Lemma 2.1.6. Assume that S : X → R has finitely many non-degenerate critical points
x1 , . . . , xN and around each critical point there are coordinates y 1 , . . . , y n in which µ =
µxk dy 1 ∧ . . . ∧ dy n (with µxk constant). Then
Z N Z
i i i i j i
X
I(~) = e ~
S
µ= µ xk e ~
S(xk )
dn ye 2~ (Hxk S)ij y y e ~ P (y) + O(~∞ ) (2.9)
X k=1 Rn

Proof. We just repeat the proof of Theorem 2.1.1 with N = ∞. The Taylor series of
g = ρ(y)ψα (y)ei/~P (y) depends only on P , since both ρ and ψα are constant in a neighborhood
of xk (again, P is the degree ≥ 3 part of the Taylor series of S at xk ).

Expanding ei/~P (y) in a power series, we are hence led to study the Fresnel moments
Z
i i j
dn ye 2~ Qij y y yi1 . . . yin .

The standard way to do this is to introduce the generating function

Z
i i j i
Q
Z [J] := e 2~ Qij y y +Ji y dn y (2.10)
Rn

(we often write just Z[J] if Q is understood from the context) which has the property that
Z
i i j ∂ ∂
dn ye 2~ Qij y y yi1 . . . yin = ··· Z[J] (2.11)
∂Ji1 ∂Jin J=0
Also, Z[J] can be explicitly computed by completing the square:
~ −1 )ij J
Z[J] = Z[0]e 2i (Q i Jj
(2.12)

and we know Z[0] from Lemma 2.1.2. We thus arrive at the following proposition:

Proposition 2.1.7. Denote Hxk S =: Qk , Zk [J] =: Z Qk [J] and Pk (y) the terms of degree 3
and higher in the Taylor series of S. With the assumptions of Lemma 2.1.6, we then have
N
i ∂ Zk [J]
µxk e ~ S(xk ) Zk [0]e ~ Pk ( ∂J )
i
X
I(~) '~→0 + O(~∞ ) (2.13)
J=0 Zk [0]
k=1

44
Remark 2.1.8. A priori, this series depends on the choice of coordinates around the critical
points, because the higher derivatives of the action do. However, one can show that this
dependence cancels out in the sum over all diagrams at every order [Joh10].
Remark 2.1.9. The limit ~ → 0 is known as the “semiclassical” limit in physics, whereas
“perturbative” usually refers to taking the coupling constant(s) to 0. At least in the case
where there is a single coupling constant, the two expansions are equivalent, as one can see
√
by rescaling the fields with ~.

2.1.3 Feynman diagrams

To label the terms in (2.13), the asymptotic expansion of oscillatory integrals to all orders
in ~, it is convenient to introduce Feynman diagrams. We try to give a self-contained but
slightly condensed introduction here, but there are plenty of excellent sources in the liter-
ature. A very pedagogical introduction for mathematicians is [Pol05]. In this exposition
we follow closely [Res10], [Mne17, Chapter 3]. In physics the use of Feynman diagrams is
usually derived somewhat differently,3 but the outcome is completely equivalent. Mathemat-
ically, Feynman diagrams label the terms appearing in Gaussian (or Fresnel) moments. The
physical interpretation is that they represent processes that happen between particles. The
corresponding Gaussian (or Fresnel) moment is interpreted as the probability amplitude of
that process.
i ∂
We thus set out for a graphical representation of e ~ Pk ( ∂J ) Zk [J]. We briefly introduce
J=0
some combinatorial terminology:

Definition 2.1.10. Let I be a finite set.

• A partition of I is a collection P = {I1 , . . . , In } of pairwise disjoint subsets of I such

that I = nk=1 Ik . The set of all partitions of I is denoted PI . The set of all partitions
S

of {1, . . . , n} is denoted Pn .

• A perfect matching m on I is a partition of I into two-element subsets. The set of

all perfect matchings is denoted MI . The set of all perfect matchings of {1, . . . , n} is
denoted Mn .
3
See e.g. [PS95] or any other QFT textbook for an account of this.

45
A perfect matching on I is given by m = {{a1 (m), b1 (m)}, . . . , {an (m), bn (m)}} where ai (m) 6=
bi (m), {ai (m), bi (m)} ∩ {aj (m), bj (m)} = ∅ and I = nj=1 {aj (m), bj (m)}. Notice that Mn is
S

empty if n is odd. A central step is the following observation which is often called Wick’s
Lemma.

Lemma 2.1.11 (Wick’s Lemma).


∂ ∂ 0 n odd
··· Z[J]/Z[0] =
∂Ji1 ∂Jin J=0  ~ )
m P −1 ia1 (m) ib1 (m)
. . . (Q−1 )ial (m) ibl (m)
i m∈Mn (Q ) n = 2m
(2.14)

~ −1 )ij J
Proof. If n is odd, the claim follows from noticing that Z[J]/Z[0] = e 2i (Q i Jj
does not
contain terms of odd orders. If n = 2m, we realize that the only surviving term is the order
m term in the exponential series
 m
∂ ∂ ∂ ∂ ~ 1
··· Z[J]/Z[0] = ··· ((Q−1 )ij Ji Jj )m
∂Ji1 ∂Jin J=0 ∂Ji1 ∂Jin J=0 2i m!
 m
∂ ∂ ~ 1
= ··· (Q−1 )k1 l1 Jk1 Jl1 · · · (Q−1 )km lm Jkm Jlm
∂Ji1 ∂Jin J=0 2i m!

A term in this sum survives if and only if every derivative can be matched to a J, that is there
is a permutation σ ∈ Sn such that (iσ(1) , . . . , iσ(n) ) = (k1 , l1 , . . . , km , lm ). Such a permutation
defines a perfect matching m = {{(σ(1), σ(2)}, . . . , {σ(n−1), σ(n)}}. Two permutations give
the same term in the sum precisely if they correspond to the same matching. The claim now
follows from the observation that the same matching appears 2m m! times (we can exchange
the two elements of a pair and permute the pairs among themselves).

Thus we have explained how every term in ei/~P (∂/∂J) acts on Z[J]/Z[0]. To go one step
further, we expand the formal power series P (remember it starts in degree 3)
∞
1 1 X 1
P (y) = Pijk y i y j y k + Pijkl y i y j y k y l + . . . = Pi1 i2 ...ik y i1 · · · y ik (2.15)
3! 4! k=3
k!

where all P ’s are symmetric in all indices. We now expand the exponential using the multi-

46
nomial theorem as
∞  N
X i 1
exp(i~P (∂/∂J)) = P (∂/∂J)N
N =0
~ N!
∞   N X  N  1 j3  jl
X i 1 1
= P3 (∂/∂J) ··· Pl (∂/∂J)
~ N ! j1 · · · jl 3! l!
N =0 j3 +...+jl =N
∞ ∞   P ji
X X i 1
=
l=3 j ,...,j =0
~ j3 !(3!) · · · jl !(l!)jl
j 3
3 l

∂ ∂
Pi1 i2 i3 · · · Pi3j3 −2 i3j3 −1 i3j3 · · · Pin−l+1 ...in ··· , (2.16)
∂Ji1 ∂Jin
Pl
where in the last line we defined n = j=1 nj j and enumerated the vertices accordingly.
We want to apply Wick’s Lemma 2.1.11. To keep track of the corresponding terms one can
introduce graphs. We give here a definition adapted to our needs.

Definition 2.1.12. • A graph Γ = (H, V, E) consists of a finite set H together with a

partition V of H and a perfect matching E of H.

• Two graphs Γ = (H, V, E), Γ0 = (H 0 , V 0 , E 0 ) are isomorphic if there is a bijection

ϕ : H → H 0 with ϕ(V ) = V 0 and ϕ(E) = E 0 .

Remark 2.1.13. Let H = {1, . . . , n}, and Γ = (H, V, E) be a graph. Suppose V has nj blocks
P
of size j = 1, . . . l, n = nj j. Then Γ is isomorphic to a graph where V is the standard
partition P0 with nj blocks of size j, i.e. P0 = {{1}, . . . , {n1 }, {n1 + 1, n1 + 2}, . . . , {n − l +
1, . . . , n}}.

We introduce some further terminology. The set H is called the set of half-edges, V is called
the set of vertices, and E is called the set of edges.

Next, we consider the automorphism group Aut(Γ) of a graph. To this end, we first note that
the symmetric group Sn acts on the set Pn of partitions of {1, . . . , n}, stabilizing the number
and size of blocks. The stabilizer subgroup (Sn )P of a partition4 P ∈ Pn with nj blocks of
size j is isomorphic to j Snj n (Sj )nj ⊂ Sn . The following proposition is immediate from
Q

the definition of graph isomorphism:

Proposition 2.1.14. Let Γ = (H, V, E) be a graph. Then the automorphism group of Γ is

the stabilizer group of the pair (V, E) under the diagonal action of SH .
4
Two partitions of a set I are the same if they coincide as sets of subsets of I.

47
 i ∂
We are now ready to give an expression for e ~ P ( ∂J ) Z[J]
Z[0]
. To this end we define the
J=0
Feynman weight of a graph.

Definition 2.1.15. Let W be a vector space and (Pk )∞

k=1 be a family of symmetric tensors
Pk ∈ Symk W ∗ and K ∈ Sym2 W . Let Γ = (H, V, E) be a graph. Let L = {l : H →
{1, . . . , |H|}, l bijective} be the set of all labeling of the half-edges of Γ. Choose a basis
of v1 , . . . vr of W with coordinate functions x1 , . . . , xr and expand Pk = Pi1 ...ik xi1 ...ik and
K = K ij vi vj .5 Then we define
X Y Y
F P,K (Γ) := Pl(h1 )...l(h|v| ) K l(h1 )l(h2 ) ≡ F (Γ) (2.17)
l∈L v={h1 ,...,h|v| }∈V e={h1 ,h2 }∈E

(here by ≡ we mean that the superscripts P, K are to be understood from the context).

Clearly F (Γ) is independent of the chosen basis because it is given by a contraction of certain
tensors. The following Proposition follows from the fact that the tensors Pk are symmetric:

Proposition 2.1.16. If Γ is isomorphic to Γ0 , then F (Γ) = F (Γ0 ).

The main theorem of this section is the following:

Theorem 2.1.17. If P is a formal power series on V ∗ starting in degree 3, and Q is a

non-degenerate symmetric bilinear form on V , then

i
P ∂
( ∂J ) Z Q [J] X (−i~)−χ(Γ) P,Q−1
e ~ = F (Γ). (2.18)
J=0 Z Q [0] |Aut(Γ)|
[Γ]

Here the sum goes over isomorphism classes of graphs that are at least trivalent (all the blocks
in the partition V have size at least 3), we use the components of P to define symmetric
tensors Pk as in (2.15), and χ(Γ) is the Euler characteristic χ(Γ) = V (Γ) − E(Γ).

Proof. We start with the expression in Equation (2.16) and apply Wick’s Lemma (2.1.11).
P
Every term in the sum is labeled by a standard partition P0 [n] of n = jnj with nj blocks
of size j. By the Wick Lemma, we get a sum over all matchings of [n] = {1, . . . , n}. The
term corresponding to the perfect matching m is precisely F (Γ), where Γ = ([n], P0 [n], m).
By Remark 2.1.13, we obtain all isomorphism classes of graphs in this way. Notice that they
5
We remind the reader that repeated indices are summed over.

48
appear with the correct power (−i~)|E|−|V | of −i~. The only question left is the combinatorial
factor. The stabilizer group of the standard partition (Sn )P0 ⊂ Sn acts on graphs of the form
([n], P0 [n], m) by acting on the perfect matching m. It satisfies |(Sn )P0 | = nj !(j!)nj ! and
Q

graphs isomorphic to ([n], P0 [n], m) are precisely given by ([n], P0 [n], m0 ) with m0 in the m-
orbit (Sn )P0 m of this action. Thus,

i ∂ Z Q [J] (−i~)1−χ(Γ)
e ~ P ( ∂J )
−1
X
= |(Sn )P0 · m|F P,Q (Γ)
J=0 Z Q [0] |(Sn )P0 |
[Γ=([n],P0 ,m)]
X (−i~)1−χ(Γ) −1
= F P,Q (Γ),
|Aut(Γ)|
[Γ]

where the last equality follows from the orbit-stabilizer theorem6

(Sn )P0
|Aut(Γ)| = |((Sn )P0 )m | = .
|(Sn )P0 |

2.1.4 Feynman graphs and rules

Above we have discussed graphs describing the asymptotic behavior of the partition function.
The vertices and edges were all indistinguishable. One can easily extend this discussion to
keep track of different terms in the quadratic operator by “decorating” the graphs. Automor-
phisms of graphs then have to be replaced by automorphisms of decorated graphs. Loosely,
one says that one computes a quantity “by Feynman graphs and rules”: One just specifies
the different types of vertices and edges appearing in the graphs - the possible graphs are
then called the “Feynman graphs”. One specifies how to compute the Feynman weight of a
graph by specifying it on the generators - these prescriptions are called the “Feynman rules”,
and they can be read off from the action functional. This often provides an elegant way of
generating all the terms in a complicated expression such as

i ∂ Z Q [J]
e ~ P ( ∂J ) ,
J=0 Z Q [0]

or other quantities of interest.

6
The author is grateful to [Link] for pointing this out.

49
2.2 Oscillatory integrals with degenerate phase func-
tions and Faddeev-Popov method

Many action functionals that appear in physics actually do not have non-degenerate critical
points, including the one that forms the main focus of these lectures, the Chern-Simons ac-
tion functional. In that case, the Hessian at the critical point has a kernel and one can ask if
the vectors in the kernel can be extended to symmetries of the action. In that case, there is
a (local or global) distribution of symmetries V ⊂ T M , i.e. all vector fields V tangent to V
annihilate the action: LV S = 0. In general, V may fail to be integrable. However, in many
examples, this distribution is not only integrable, but actually comes from a group action on
the space of fields preserving the action functional. In these cases one can use the Faddeev-
Popov method to analyze the behavior of the oscillatory integral. See also [Pol05],[Res10] or
[Mne17, Section 4.1]

2.2.1 Setup

Again, we consider the integral Z

I(~) = ei/~S µ (2.19)
X
with notations and assumptions as in Theorem 2.1.1. We now assume that there is a compact
Lie group G of dimension k which acts freely (i.e. with trivial stabilizers) on X such that
S and µ are G-invariant. We denote this action by ρ : G × X → X, ρ(g, x) = g · x, and the
corresponding infinitesimal action by ρ# : g → Γ(T M ), i.e.

d
ρ# (ξ)x = exp(tξ) · x ∈ Tx X. (2.20)
dt t=0

We frequently denote ρ# (ξ)x = ρ# #

x (ξ), this defines, for all x ∈ X, a map ρx : g → Tx X. For a
fixed basis Ta , a = 1, . . . , k of g, we denote va = ρ# (Ta ) the corresponding fundamental vector
fields. The condition that µ is invariant means that Lva µ = 0 for a = 1, . . . , k. We denote
the quotient map by p : X → X/G, since G is compact and the action is free, X/G is also a
manifold, and the action descends to a smooth function S̃ ∈ C ∞ (X/G). The dimension of
the quotient is n − k =: l. We denote µ̃ = p∗ µ the pushforward (integral along the fiber of

50
p) of µ. Then, we have Z Z
i i
S
e ~ µ= e ~ S̃ µ̃. (2.21)
X X/G

By invariance of µ we have ιvk · · · ιv1 µ = Cp∗ µ̃ for some constant C, we assume µ is normalized
7
in such a way that C = 1. The integral over the quotient now admits an asymptotic
expansion if the following assumption is satisfied:
Assumption 2.2.1. The critical orbits of S are isolated and S̃ has a non-degenerate Hessian
at these orbits.

Even if this assumption is satisfied, we still often do not have a good way to think of the
quotient and S̃, µ̃ (think e.g. of the action of gauge transformations on connections). The
Faddeev-Popov method gives an alternative way to compute the integral by means of more
tractable data.

2.2.2 Gauge fixings

Let x0 be a critical point of S. We say that a chart (U ; y 1 , . . . , y l , z 1 , . . . , z k ) is adapted if

 2 l
∂ S
∂yi ∂yj
is non-degenerate and va = fab ∂z∂ b .
i,j=1

We define a map φ : U → g by φ(y, z) = z b Tb . It follows that φ defines a local section s of

the quotient by s(x) = [x] ∩ φ−1 (0). Such a section is known as a (local) gauge-fixing. φ is
called the gauge-fixing function.
Assumption 2.2.2. There is a global function φ : X → g (a global gauge-fixing function) such
that φ−1 (0) intersects every orbit of G transversally exactly N times.

Very often it will not be possible at all to find functions φ with N = 1: Consider e.g. the
example of the circle acting on the cylinder I × S 1 by rotation. Any global gauge-fixing will
intersect every orbit at least twice (exercise).
Given such a global gauge-fixing, we can rewrite the integral over the quotient as an integral
over a subset of X employing a delta function:
Z Z
i 1 i
I(~) = S
e ~ µ̃ = δ (l) (φ)p∗ µ̃ e ~ S . (2.22)
X/G N X
7
This explains a missing factor of vol(G) in comparison with [Mne17, Section 4.1]

51
Here δ (l) (φ) = δ(φ)dφ1 ∧ · · · ∧ dφk . Our next goal is to rewrite the integral in terms of the
original measure µ, i.e. find a function8 J such that dφ1 ∧ · · · ∧ dφk p∗ µ̃ = Jµ.

Lemma 2.2.3. J is given by J(x) = det F P (x) where F P (x) is the Faddeev-Popov operator

F P (x) : g → g
ξ 7→ dx φ(ρ#
x ξ)

In the basis Ta , the Faddeev-Popov operator is given by F P (x)ab = dφa (vb (x)).

Proof. Extend dφ1 , . . . , dφk to a basis of T ∗ M by α1 . . . , αl s.t αj (vk ) = 0 and µ = dφ1 ∧

· · · ∧ dφk ∧ α1 ∧ · · · ∧ αl . Then, one simply computes
X
ιvk . . . ιv1 µ = (−1)sgn(σ) dφ1 (vσ(1) ) · · · dφk (vσ(k) ) ∧ α1 ∧ · · · ∧ αk
σ∈Sk

= det F P (x)α1 ∧ · · · ∧ αk .

Wedging with dφ1 ∧ · · · ∧ dφk , we obtain

dφ1 ∧ · · · ∧ dφk ιvk . . . ιv1 µ = det F P (x)µ

which proves the claim since ιvk . . . ιv1 µ = p∗ µ̃.

We can use this Lemma in order to rewrite the integral in terms of µ:

Corollary 2.2.4. Z
1 i
I(~) = µ δ(φ(x)) det F P (x)e ~ S (2.23)
N X

Our next goal is to rewrite this as the integral of an exponential so that we can again use
the Feynman diagrammatic methods in the last Section. We can rewrite the Delta function
as a Fourier transform Z
1 i
δ(φ(x)) = e ~ hλ,φ(x)i dm λ. (2.24)
2π~ λ∈g∗

What about the determinant? The main realization in the Faddeev-Popov formalism (some-
times called the Faddeev-Popov trick) is that it can be written as an integral of an exponential
by introducing “odd” coordinates, which we shall briefly introduce in the next section.
8
The Radon-Nikodym Derivative of the corresponding measures.

52
2.2.3 Super vector spaces and Berezin integrals

Berezin integrals, introduced by Berezin [Ber66], are integrals over super vector spaces (see
also [CS11] for a concise introduction, for a more comprehensive introduction to superge-
ometry and symmetry from the mathematical viewpoint, see [Var04]). Those, in turn, are
combinations of even and odd vector spaces - even vector spaces are just the usual ones,
whereas odd vector spaces are defined through their algebra of functions.

Odd and super vector spaces

Definition 2.2.5. Let V be a vector space over R, then we define9 the odd vector space ΠV
by O(ΠV ) = ∧• V ∗ . Here O denotes the algebra of functions.

The upshot is that the coordinates ξ 1 , . . . , ξ n on ΠV anticommute

ξiξj = ξj ξi. (2.25)

These are known as Grassmann, fermionic or simply odd coordinates.

A super vector space is now simply the direct sum of an even and an odd vector space.

Definition 2.2.6. A super vector space V is the direct sum of an even and an odd vector
space: V = V0 ⊕ ΠV1 . Its algebra of functions is O(V) = C ∞ (V0 ) ⊗ O(ΠV1 ).

Remark 2.2.7. Below we will also encounter objects of the form X = X × ΠV , where X is
a manifold. Those are basic examples of supermanifolds, and since they are the only ones
relevant for this text, we refrain from giving the somewhat lengthy technical definition. Their
main property is that O(X ) = C ∞ (X) ⊗ O(ΠV ).

Derivatives

The algebra of functions on an odd vector space is a super algebra, i.e. a Z2 -graded algebra.
The grading is given by the degree in the ξ’s (modulo 2) and derivatives are super-derivations
9
Purists might prefer to say that the category of odd vector spaces is by the definition the opposite
category of the category of free finite-dimensional anticommutative algebras.

53
of this super-algebra, i.e. ∂/∂ξi is defined by

∂ ∂ ∂
i
(f g) = i f g + (−1)|f | i g (2.26)
∂ξ ∂ξ ∂ξ
∂ j
ξ = δij (2.27)
∂ξ i

Integrals

To define integrals we first define them on the “odd line” ΠR. Here the integral is completely
defined by the two requirements
Z
∂g
Dθ =0 (2.28)
ΠR ∂θ
Z
Dθθ = 1. (2.29)
ΠR

The algebra of functions on the odd line is O(ΠR) = R + Rθ. The integral of a function
f = a + bθ is then simply Z
Dθf = b.
ΠR

Warning: The symbol Dθ is not a one-form. In fact, if we define a new coordinate θ0 = λθ,
then Equation (2.29) implies
Z Z
0 0
1= Dθ θ = Dθ0 λθ
ΠR ΠR

and thus Dθ0 = λ1 Dθ (in contrast, one-forms transform as dθ0 = λdθ!). To define the integral
on ΠRn , we iterate Z Z Z Z 
n
D θ := ··· Dθ1 .
ΠRn ΠR ΠR ΠR

An easy exercise shows that this implies

Z
Dn θf = f top (2.30)
ΠRn

where f top is the component of f that lies in ∧n Rn . This is turn means that in the odd case,
integral and derivative coincide:
Z
∂ ∂
Dn θf = ... f.
ΠRn ∂θn ∂θ1

54
Equation (2.30) suggests how to define the integral in an abstract odd vector space ΠV :
Namely, by the choice of an identification ∧top V ∗ ∼
= R. Such an identification is equivalent
to an element µ ∈ ∧V ∗ (called a Berezinian) and we can define the integral
Z
µf = hµ, f i. (2.31)
ΠV

In particular, the odd vector space Π(V ⊕ V ∗ ) has a canonical Berezinian given by

µc = Dθn Dθ̄n · · · Dθ1 Dθ̄1 (2.32)

for any choice of coordinates θi with dual coordinates θ̄i on ΠV ∗ . The crucial property of the
canonical Berezinian is that for an endomorphism B ∈ End(V ) ∼ = V ∗ ⊗ V ⊂ O(ΠV ⊕ ΠV ∗ )
we have Z
j i
det B = µc eBi θ θ̄j . (2.33)
ΠV ⊕ΠV ∗

We have thus completed the task we set out for and found a representation of the determinant
as the integral of an exponential!

2.2.4 The Faddeev-Popov action

Equations (2.24) and (2.33) allow us to rewrite the integral (2.23) as the integral of an
exponential. To this end, we collect the Faddeev-Popov data we have constructed so far.

Definition 2.2.8 (Faddeev-Popov data). Let X be a manifold with a free Lie group action
ρ : G × X → X, and S : X → R a G-invariant function, µ ∈ Ωtop (X) a G-invariant volume
form. Suppose we are given a global gauge-fixing function φ : X → g. Then we define

• The Faddeev-Popov space of fields

XF P = X × g∗ × Π(g ⊕ g∗ ) 3 (x, λ, c, c̄) (2.34)

λ is called the Lagrange multiplier and (c, c̄) the ghost-antighost pair.

• For every x ∈ X the Faddeev-Popov operator F P (x) : g → g by

F P (x) = dφ(x) ◦ ρ#
x (2.35)

55
 • The Faddeev-Popov action SF P ∈ O(XF P ), defined by

SF P (x, λ, c, c̄) = S(x) + hλ, φ(x)i + hc̄, F P (x)ci (2.36)

• The Faddeev-Popov Berezinian µF P by

1
µF P = µ dn λ µcΠ(g⊕g∗ ) (2.37)
N (2πi)m

where µcΠ(g⊕g∗ ) is the canonical Berezinian on Π(g ⊕ g∗ ) introduced in (2.32).

The following is one of the two main results of this section:

Theorem 2.2.9. In the notation as above, we have

Z Z
i i
S
e µ=
~ e ~ SF P µF P (2.38)
X XF P

Proof. This follows from Corollary 2.2.4 together with equations (2.24) and (2.33).

Critical Points

The main point of the Faddeev-Popov action (2.36) is that - in contrast to the action S -
its critical points are non-degenerate, at least under the assumption that the FP operator is
non-degenerate.10

Proposition 2.2.10. Assume that F P (x) is non-degenerate for all x. Then the critical
points of SF P (x) are given by 
c =0






c̄ =0





φ(x) =0 (2.39)






 λ =0


dx S

=0

I.e. Crit(SF P ) = {Crit(S) ∩ φ−1 (0)} × {(0, 0, 0)} ⊂ XF P .

10
In the finite-dimensional case this is equivalent to φ−1 (0) intersecting the G-orbits transversally.

56
Proof. The first three equations follow (using non-degeneracy of F P ) by taking derivatives
with respect to c̄, c and λ respectively. For the last equation, notice that taking derivative
with respect to x yields
dx S + hλ, dx φi = 0.
Solutions of this equation are extrema of S under the condition φ = 0. By G-invariance of
S, such conditional extrema are also global extrema and hence dS = 0 at these points. It
follows that also λ = 0.
Proposition 2.2.11. Assume F P (x) is non-degenerate for all x ∈ X. Then the critical
points of SF P are non-degenerate.

Proof. To see this one computes the Hessian of SF P with respect to an adapted chart x =
(y, z) in X at a critical point x0 ≡ ((y0 , 0), 0, 0, 0) ∈ XF P :
 2S

( ∂y∂i ∂y l
j )j=1 0 dy φT 0 0
 

 0 0 dz φT 0 

Hx0 SF P =  dy φ dz φ 0 0 0 (2.40)
 

 

 0 0 0 0 F P (x0 )T 

0 0 0 F P (x0 ) 0
In adapted coordinates, we have φ(y, z) = z, whence we conclude that dy φ = 0 and dz φ =
2
Id. By assumption, ( ∂y∂i ∂y
S l
j )j=1 and F P (x0 ) are non-degenerate. Hence Hx0 SF P is non-
degenerate.

2.2.5 Stationary phase method for Faddeev-Popov integrals

Since we have replaced the degenerate function S with the non-degenerate SF P , our next
aim is to generalize the methods developed in the last section to the case at hand. The main
difference is that the determinant of the odd quadratic operator appears in the nominator
(rather than the denominator) of the prefactor. Let us first write the Hessian of SF P at x0
and its inverse without adapted chart:
2S
 
( ∂x∂i ∂x n
j )j=1 dx φT 0 0
 
 dx φ 0 0 0 
Hx0 SF P =   (2.41)

 0 0 0 F P (x0 )T 

0 0 F P (x0 ) 0

57
  
K γT 0 0
 
−1
 γ 0 0  0
(Hx0 SF P ) =  (2.42)
 0
 0 0 (F P (x0 )−1 )T 

−1
0 0 (F P (x0 ) ) 0
−1 −1
Here γ = ρ#
x0 ◦ F P (x0 ) . We denote by Qx0 the Hessian of S restricted to φ (0). We can
then express the asymptotic behavior of I(~) as
i
(n−k)/2
X i
S(x0 ) e ~ sign Qx0 X (−i~)1−χ(Γ)
I(~) '~→0 vol(G)(2π~) e ~ det F P (x 0 ) F (Γ)
| det Qx0 |1/2 Γ
|Aut(Γ)|
crit. G-orbits [x0 ]
(2.43)
Here the sum goes still over Feynman graphs, but the graphs can have different generators
(in the spirit of subsection 2.1.4), which we list below. The proof of this equality is very
similar to the one given in the purely even case, the only difference are that one has to
compute momenta of odd Gaussians. The main difference is that the determinant prefactor
appears with a positive power. We leave the details to the reader as an exercise. The graphs
are generated by the half-edges in Table 2.1 below.

coordinate half-edge name

xi i “field(s)”
j
λj Lagrange multiplier11
j
cj ghost
c̄j i antighost

Table 2.1: Half-edges in FP Feynman diagrams, i runs from 1 to n, whereas j runs from 1
to k.

Edges are perfect matchings on the space of half-edges. Note that that the Feynman rules do
not associate edges (as opposed to vertices, see below) with individual terms in the action,
but rather, edge types that are not listed below evaluate to 0 (the corresponding block in
the inverse of the Hessian vanishes).
11
In physics literature, one often uses another version of the Faddeev-Popov formalism where λ is not
needed, if it is introduced, it is sometimes called “Nakantishi-Lautrup field” (e.g. [Wei05]).

58
 edge operator name

i j
K ij propagator in gauge φ = 0
i j
γij “ x-λ propagator”
i j
F Pji ghost propagator

Table 2.2: Edges in FP Feynman diagrams and their corresponding operators

Vertices are given by a partition of the set of half-edges. Under the Feynman rules, like in
the case before, they are given by third and higher degree derivatives of the action. Every
term in the action yields a corresponding vertex. Since the action is linear in λ, c, and c̄,
there is at most one of the corresponding half-edges at every vertex.12 See Table 2.3.

vertex tensor name

i2
∂s
∂xi1 ...∂xis
S(x) x=x0 , s ≥3 x s-vertex13
is i1
i2

il i1 ∂s
φj (x) x=x , l ≥2 λ l-vertex
∂xi1 ...∂xil 0

j
i2

im i1 ∂s
F Pjj21 (x) x=x , m ≥1 ghost m-vertex
∂xi1 ...∂xim 0

j1 j2

Table 2.3: Vertices in FP Feynman diagrams and their corresponding tensors

Mathematically, the graphs are modeled by tuples (H = He ∪Ho , V, me , mo ) where He denotes

12
Again, although in principle there are of course graphs with e.g. 5 λ half-edges at one vertex, they
evaluate to 0 under the Feynman rules, since the corresponding derivative of the action vanishes. Usually
one does not count them among the Feynman graphs appearing in the theory.
13
In gauge theories the field (here represented by x) usually has a particular name (“photon” in QED,
“gluon” in QCD, etc.) and the vertex is then called after the field.

59
the set of even and Ho the set of all odd half-edges. me , mo are perfect matchings on He , Ho
respectively. The additional structure in the Feynman rules above (no λ − λ edges, only
directed c̄ − c edges) comes from the vanishing of the corresponding block in the Hessian,
i.e. F (e) = 0 on these unwanted edges, F (v) is automatically 0 on vertices that were not
listed above because the corresponding derivatives of SF P vanish.
Remark 2.2.12. There are two special cases that simplify the diagrams in question. In
examples, the gauge-fixing function can often chosen to be linear in the fields. In that case,
the λ vertices with dotted edges vanish (they correspond to at least two derivatives of the
gauge-fixing function). This is the case in the covariant gauges usually applied in Yang-Mills
theories, discussed in the example below.
Another special case occurs if the FP operator is locally constant in x. In that case the
weight of the ghost vertex vanishes. This happens in abelian Yang-Mills theories (such as
QED).

2.2.6 Example: Yang-Mills

Yang-Mills theory can be used to describe the dynamics of gluons, the particles that trans-
mit the electro-magnetic and weak and strong forces. The space of fields is the space of
connections on a trivial principal bundle M × G, where M is a Riemannian or Lorentzian
manifold and G ⊂ GL(n) a connected and compact Lie group14

FM = Conn(M × G) ∼
= Ω1 (M, g) (2.44)

with action functional Z

1
tr FA ∧ ∗FA .
SM = (2.45)
M 2
Here ∗ denotes the Hodge star and we remind the reader that FA is the curvature of the
connection A given by FA = dA + 12 [A, A]. The gauge group G = C ∞ (M, G) acts on FM via

g · A = ρ(g)A = g −1 Ag + g −1 dg (2.46)

and the infinitesimal action of the Lie algebra Lie(G) = Ω0 (M, g) is given by

ρ#
A (α) = dA α ∈ TA FM . (2.47)
14
In particle physics, G = U (1) describes the electromagnetic force, G = SU (2) the weak force and
G = SU (3) the strong force.

60
A gauge that is often chosen is the Lorenz gauge

φ(A) = d∗ A = 0 (2.48)

so that the gauge fixing function is φ = d∗ : Ω1 (M, g) → Ω0 (M, g) and the Faddeev-Popov
operator F P (A) = dφ ◦ ρ# 0 0
A : Ω (M, g) → Ω (M, g) is given by

F P (A)α = d∗ dA α.

In the infinite-dimensional setting, one has to chose an appropriate model for Ω0 (M, g)∗ . A
possible choice is Ωd (M, g), where the action of λ ∈ Ωd (M, g) on α ∈ Ω0 (M, g) is given by
Z
hλ, αi = trλ ∧ α. (2.49)
M

The space of Faddeev-Popov fields is then

FF P = FM × Ωd (M, g) × ΠΩ0 (M, g) × ΠΩd (M, g) 3 (A, λ, c, c̄) (2.50)

and the Faddeev-Popov action is

Z  
1 ∗ ∗
SF P [A, λ, c, c̄] = tr FA ∧ ∗FA + λ ∧ d A + c̄ ∧ d dA c . (2.51)
M 2

Perturbative evaluation of the path integral

Z pert X (−i~)−χ(Γ)
i
pert
Z = e ~ SF P ' F (Γ)
FF P Γ
|Aut(Γ)|

leads to a series of terms labeled by Feynman graphs. As discussed above, there are in
principle three types of edges in these diagrams but since the gauge-fixing is linear, the
“dotted” edges (or “A-λ” propagators) do not appear. See Table 2.4. The propagator
in Yang-Mills is usually denoted with a curly line, and called gluon propagator (after the
particles corresponding to excitations of the gauge field).

The vertices correspond to the cubic and higher terms in the action functional, see Table 2.5.
A major problem is that closed gluon loops will result in divergent integrals. For Yang-Mills
theory this problem has been solved by the process of renormalization, see e.g. [Hol08].

61
 edge operator name

i j
K ij gluon propagator in gauge φ = 0
i j
F Pji ghost propagator

Table 2.4: Edges in Yang-Mills diagrams in Lorenz gauge and their corresponding operators

vertex term in action functional name

1
R
2 M
tr[A, A] ∗ dA 3-gluon vertex

1
R
8 M
tr[A, A] ∗ [A, A] 4-gluon vertex

j1 j2
c̄ ∧ d∗ [A, c]
R
M
ghost vertex

Table 2.5: Vertices in FP Feynman diagrams and their corresponding tensors

2.2.7 Example: Gravity

In principle, the perturbative Faddeev-Popov formalism can be applied to gravity just in the
same way. The space of fields is the space of Lorentzian metrics on M :

FM = M et1,3 (M ) (2.52)

and the action functional (in the vacuum) is

Z
SM [g] = Rg dvolg , (2.53)
M

where Rg denotes the Ricci scalar. The gauge group is the group of diffeomorphisms of M ,
and its Lie algebra is the Lie algebra of vector fields: G = Diff(M ), Lie(G) = X(M ). The
infinitesimal action of X(M ) on FM is

2 ∗
ρ# 1,3
g (ξ) = Lξ (g) ∈ Tg M et (M ) ' S T M. (2.54)

62
A gauge-fixing condition often used is the de Donder or harmonic coordinates gauge-fixing
given by
φ(g) = Γρµν g µν = 0,

where Γ denotes the Christoffel symbols, and we have

hc̄, F P (g)ci = g µν g ρσ c̄ρ ∂µ ∂ν cσ .

Due to the highly non-linear nature of all three terms in the Faddeev-Popov action, all
possible graphs discussed above can appear (with any valence of “graviton” legs). For this
theory, no satisfactory renormalization procedure has yet been found.15 .

2.3 BRST symmetry

As already remarked in Henneaux and Teitelboim [HT94], a curiosity in gauge-fixing is that

to “remove” unwanted degrees of freedom, one adds extra variables. An explanation of this
phenomenon is the mathematical concept of resolutions. The BRST symmetry (discovered
independently by Becchi, Rouet and Stora in [BRS75], [BRS76] and Tyutin in [Tyu76]) is the
starting point of the connection between gauge theory and homological (or derived) algebra.

2.3.1 BRST operator

Remember that
XF P = X × g∗ × Π(g ⊕ g∗ ) 3 (x, λ, c, c̄)

from where we have

O(XF P ) = C ∞ (X) ⊗ Ŝg ⊗ ∧• g∗ ⊗ ∧• g.
c
Again let Ta be a basis of g, fab the corresponding structure constants and ρ# (Ta ) = vai ∂x∂ i .
O(XF P ) is a super algebra (the Z2 -grading comes from the exterior algebras) and there is
an odd derivation Q on this superalgebra (put differently, an odd vector field on XF P ):
15
Which does not mean that it does not exist. See, for instance, the recent discussion in [Kre08], [Pri18],
and references therein.

63
Definition 2.3.1. The BRST operator or BRST symmetry is the derivation Q : O(XF P ) →
O(XF P ) defined on generators by

Qxi = ca vai (x) (2.55)

1 c a b
Qcc = fab c c (2.56)
2
Qc̄a = λa (2.57)
Qλa = 0 (2.58)

The connection to homological algebra comes from the following Lemma:

Lemma 2.3.2. The operator Q : O(XF P ) → O(XF P ) squares to 0.

Proof. This follows immediately from the definition using the Jacobi identity and the fact
that ρ# is a homomorphism of Lie algebras.

2.3.2 BRST cohomology

Thus O(XF P ) is a (Z2 -graded) complex when equipped with Q. The following observation
is obvious from the definition.

Proposition 2.3.3. The complex (O(XF P ), Q) splits as a direct sum of two subcomplexes
(O(XF P ), Q) = (O(Xmin ), QCE ) ⊕ (O(Xaux ), Qaux ), where

Xmin = X × Πg (2.59)
∂ 1 c ∂
QCE = ca vai (x) i
+ ca cb fab (2.60)
∂x 2 ∂cc
(2.61)

and

Xaux = g∗ ⊕ Πg∗ (2.62)

∂
Qaux = λa (2.63)
∂c̄a

This allows us to compute the cohomology of the BRST operator:

64
Proposition 2.3.4. The cohomology of the BRST operator is

C ∞ (X/G) k=0
H k (O(XF P ), Q) = H k (O(Xmin ), QCE ) = (2.64)
0 k=1

Proof. The first equality follows from the fact that the cohomology of Qaux is trivial. The
second equality follows from the fact that H k (O(Xmin ), QCE ) is the Chevalley-Eilenberg
complex for the Lie algebra action of g on C ∞ (X).

Thus both functions on XF P and Xmin are resolutions of the quotient X/G. Since the
“physical” observables are elements of the quotient we interpret the even cohomology of the
BRST operator as the physically relevant information.

2.3.3 Gauge fixing

It follows from gauge invariance of S that S is Q-closed. From the definition of Q, the
following Lemma is immediate:

Lemma 2.3.5. We have

SF P = S + Qψ, (2.65)

where
ψ(x) = hc̄φ(x)i (2.66)

is called the gauge fixing fermion.

Thus the gauge fixing does not change the cohomology class of the physical action.
Remark 2.3.6. It should be noted that if we are only interested in a resolution of the gauge
symmetries, Xmin suffices. However, to implement the gauge-fixing condition, we have to
add the acyclic complex Xaux .

One can promote the “BRST logic” (try to resolve the action of the gauge symmetries on
X) to a gauge-fixing formalism, and there is also a quantum version of it. However, we will
head straight to the BV formalism, and refer to the literature (e.g. [HT94]) for a deeper
discussion of BRST concepts.

65
2.4 Batalin-Vilkovisky formalism and effective actions

The Batalin-Vilkovisky formalism, introduced by Batalin and Vilkovisky in the early ’80s
([BV81],[BV83]) is the most powerful gauge-fixing formalism on the market. Its main ad-
vantages over the BRST formalism are

• it can deal with situations where the distribution of symmetries is not integrable (this
happens mostly in string theory)

• it has a natural framework for renormalization and effective actions.

The very rough idea is that - in a situation where one cannot define the perturbative integral
even on XF P - one adds, for all fields φ ∈ XF P an “antifield”16 φ+ and the perturbs the
region of integration away from the zero section XF P = {φ+ = 0} to a new region L:

φ+

In this article we discuss a simple version of the BV formalism that uses only Z2 -grading and
super vector spaces. This avoids the technical complications of graded manifolds. There are
plenty of resources on BV formalism, the author has found the introductory texts [Fio03] and
[Cat19] and references therein helpful. In these notes we follow very roughly the exposition
in [Mne17, Chapter 4].

2.4.1 Odd symplectic vector spaces

We begin with the discussion of odd symplectic vector spaces.

16
The terminology is standard but somewhat unfortunate, given that ”anti” can have many different
meanings. In particular, the antifield of the ghost c+ is not the same as the antighost c̄.

66
Definition 2.4.1. An odd symplectic vector space (or BV vector space) (V, ω) is a super
vector space V = V0 ⊕ ΠV1 with a non-degenerate bilinear pairing ω : V0 × V1 → R.

Let us explain how one can interpret the pairing ω as an odd symplectic form on V . One
can extend ω to a bilinear form on V by defining it to vanish on V0 and V1 . The extended
ω is then automatically antisymmetric in the Z2 -graded sense. It is odd since it pairs only
even and odd vectors. On a vector space, any nondegenerate antisymmetric bilinear form is
symplectic (dω = 0 since ω is constant).

Definition 2.4.2. A subspace L ⊂ V of an odd symplectic vector space (V, ω) is called

Lagrangian if ω L
= 0 and dim L = 21 dim V .

For example, V0 and ΠV1 are Lagrangian subspace of V (non-degeneracy of ω forces dim V0 =
dim V1 = 12 dim V ). In coordinates y i on V0 and θi on V1 the odd symplectic form reads

ω = dy i ∧ dθi .

We can also choose more general Darboux coordinates z = (xi , x+ i

i ) where the x can be even
or odd, as long as x+ i
i have opposite parity form x .

2.4.2 Odd Poisson bracket and BV algebra

The odd symplectic form induces an odd Poisson bracket (·, ·) : O(V ) × O(V ) → O(V ). In
Darboux coordinates z α = (xi , x+
i ) it reads

∂r f ∂l g ∂r f ∂l g
(f, g) = + i
− (−1)(|f |+1)(|g|+1) + i (2.67)
∂xi ∂x ∂xi ∂x
where the left and right derivatives are defined by
∂l f ∂f
α
= α (2.68)
∂z ∂z
∂r g α ∂g
α
= (−1)|z ||g| α . (2.69)
∂z ∂z
This bracket is called BV bracket, sometimes also the Buttin bracket or antibracket. There
is an odd operator ∆ : O(V ) → O(V ) called the BV Laplacian given17 in these Darboux
17
The story of the BV Laplacian is a lot subtler and more interesting than we make it appear here. See
the papers [Khu04],[Šev06] for a more detailed discussion of this aspect.

67
coordinates by18
X ∂ ∂
∆= (2.70)
i
∂x+
i ∂x
i

Lemma 2.4.3. The BV operator squares to zero: ∆2 = 0.

Proof. Since |xi | = |xi |+ + 1, we have

! ! ! !
X ∂ ∂ X ∂ ∂ X ∂ ∂ X ∂ ∂
2
∆ = + i + j
=− + j + i
= −∆2 .
i
∂xi ∂x j
∂xj ∂x j
∂xj ∂x i
∂xi ∂x

The cohomology of the ∆ operator is called the BV cohomology of V . The triple (O(V ), ∆, (·, ·))
satisfies the axioms of what is known as a BV algebra. In particular, one has

∆(f · g) = (∆f ) · g + (−1)|f | (f, g) + (−1)|f | f · ∆g. (2.71)

2.4.3 BV integral and gauge fixing independence

We now turn to the main idea of the BV formalism. Suppose that we equip V with the
1/2
Berezinian µBV = dn yDn θ. Then its square root µBV is the density that transforms with the
square root of the superdeterminant of the transition functions. If L ⊂ V is a Lagrangian,
1/2
the restriction of µBV is a density and one can define the BV integral
Z
1/2
f µBV (2.72)
L

of an element f ∈ O(V ). The importance of this integral stems from the following theorem.

Theorem 2.4.4 (Batalin-Vilkovisky [BV81; BV83], Schwarz [Sch93]). If ∆f = 0, then the

BV integral Z
1/2
f µBV (2.73)
L
•
depends only on [f ] ∈ H∆ (O(V )) and the homology class of L.
18
We are using the sign conventions of [Fio03].

68
In particular, the integral is invariant under small perturbations of the Lagrangian.

The gauge-fixing in the BV formalism is the choice of the Lagrangian submanifold L. If we

i
have a BV action functional SBV satisfying ∆e ~ S = 0, then the integral
Z
i 1/2
e ~ S µBV (2.74)
L

is independent under deformations of L. If we find a Lagrangian L such that the quadratic

part of the action SBV has non-degenerate critical points when restricted to L, then the BV
integral can be defined perturbatively.

2.4.4 Master equations

i
The functions f we are interested in are of the form f = e ~ S . The compatibility between
the BV Laplacian and the bracket (2.71) implies
 
i
S 1 i
∆e =
~ (S, S) − i~∆S e ~ S (2.75)
2
We are thus interested in action functionals S that satisfy
1
(S, S) − i~∆S = 0 (2.76)
2
Equation (2.76) is known as the Quantum Master Equation (QME). Its reduction modulo ~
is
(S, S) = 0, (2.77)
the classical Master Equation (CME). We are thus facing the following two problems:
Problem 1 (Classical BV extension). Given an action functional S : X → R, find a super-
vector space XBV and SBV ∈ O(XBV ) such that

• X ⊂ (XBV )0 and SBV X

= S.

• The CME holds: (SBV , SBV ) = 0.

Problem 2 (Quantum BV extension). Given XBV and SBV as above, find S̃ ∈ O(XBV )[[~]]
such that the QME holds.

A very simple case is when ∆S = 0. The QME is then automatically satisfied. More
background on the master equation can be found e.g. in [Sch93], [Ale+97] or [FK14].

69
2.4.5 Faddeev-Popov and BRST solutions to Classical Master Equa-
tion

If one has FP data, then one can construct a solution to the classical master equation as
follows. Suppose we are given XF P , SF P as above (for simplicity we assume that X is a
vector space). Then, define19

XBV = ΠT ∗ XF P = XF P ⊕ ΠXF∗ P 3 (z α , zα+ ) (2.78)

and
1 c +
SBV = S + Q(z α )zα+ = S + ca vai (x)x+
i + fab cc + λa c̄
a,+
. (2.79)
2
Given a gauge-fixing function φ(x) the gauge-fixing Lagrangian can be defined by
 
+ ∂ψ
L= z = α (2.80)
∂z

where ψ = hc̄, φ(x)i is the BV version of gauge-fixing fermion. One can then show that
SBV L
= SF P .
A similar construction works for the “minimal” space of BRST fields Xmin = X ⊕ Πg, one
can define
∗
XBV = Xmin ⊕ ΠXmin (2.81)

and
SBV = S + QCE (z α )zα+ . (2.82)

This satisfies the classical master equation.

Remark 2.4.5. In both cases, ∆S = 0 if the corresponding QBV = (S, ·) is divergence-free
for µBV . In the infinite-dimensional case the measure does not make sense and one has to
regularize the Laplacian to obtain the QME.

2.4.6 BV pushforward and effective actions

In some examples, Chern-Simons theory can be gauge-fixed by simply using the Faddeev-
Popov formalism, for instance this was done in [Pol05],[AS91], [Bar91]. However, in general
19 α
z is used with different meaning here from above.

70
it is not possible to find Lagrangian L inside the whole space of Chern-Simons fields with the
property that the action restricted to L has non-degenerate critical points, this is discussed
in more detail in Section 3.2. Here it is necessary to use the BV formalism, because it easily
allows for the incorporation of effective actions, as we shall explain presently.

Suppose we have a BV vector space (XBV , ωX ) with a splitting into BV vector spaces XBV =
0
YBV × YBV such that the symplectic form also splits: ωX = ωY + ωY 0 . Then also the BV
0
Laplacians split: ∆X = ∆Y + ∆Y 0 . We also assume that µX Y Y
BV = µBV µBV .

0
Definition 2.4.6. Let L be a Lagrangian in YBV . The effective action Seff ∈ O(YBV ) on
YBV induced by SBV ∈ O(XBV ) is defined by
Z
i i 0 0
Seff (y)
e ~ = e ~ S(y+y ) (µYBV )1/2 (2.83)
y 0 ∈L

The fiber integral on the right hand side is called a BV pushforward. In particular, we have
Z
i 0 i
e ~ S(y+y ) (µX
BV )
1/2
= e ~ Seff (y) (µYBV )1/2 (2.84)
y 0 ∈L

The following theorem is the generalization of Theorem 2.4.4 to BV pushforwards:

Theorem 2.4.7. 1. If S ∈ O(XBV ) satisfies the Quantum Master Equation, then so does
Seff ∈ O(YBV ), i.e.
i
∆Y e ~ Seff (y) = 0. (2.85)

0
2. Let Lt be a smooth family of Lagrangians in YBV with L0 = L. Denote Seff,t the effective
action defined by the gauge-fixing Lagrangian Lt . Then there is R ∈ O(YBV ) s.t.

d
Seff,t = ∆Y R. (2.86)
dt t=0

For the proof we refer to [CM08],[CMR17]. Thus, from a solution of the QME on XBV we
can define a family of effective actions on BV subspaces by BV pushforwards:

X Y (1) Y (2) ... Y (N ) ... (2.87)

71
The corresponding effective actions all satisfy the corresponding Quantum Master Equation.
In particular, this is a version of “Wilson renormalization”.20 In our setting (for topological
theories) we are interested in a BV pushforward from the space of fields to the space of
zero modes. These prevent a theory from being handled in the FP or BRST formalisms and
is one of the main reason for resorting to BV formalism. The other reason to use the BV
formalism is that is better adapted to manifolds with boundary and cutting and gluing, as
was recently shown in the work of Cattaneo, Mnev and Reshetikhin in [CMR14], [CMR17].
Thus one can hope to compute the BV effective action by cutting the spacetime into simple
pieces and gluing them back together. This is the focus of ongoing research, see for instance
[CMW17], [CMW18].

20
See the work of Anselmi [Ans94] on BV formalism and renormalization, or [Cos11] for a modern treatment
of the subject.

72
Section 3

Perturbative Quantization of
Chern-Simons Theory

In this final Section we turn to the perturbative quantization of Chern-Simons theory. To

construct the perturbative partition function we go through several steps. We start off
with the Batalin-Vilkovisky formulation of Chern-Simons theory. We then fix a classical
background - i.e. a flat connection A0 - and gauge fix in the neighborhood of that classical
background. Since there can be zero modes around, we then define the partition function
as the result of a formal BV pushforward to the space of zero modes. Finally, we show that
while this partition function satisfies the quantum master equation, its BV cohomology class
is not invariant under deformations of the gauge-fixing Lagrangian, as one would expect
from the finite-dimensional BV formalism. We then comment on how to fix this problem by
introducing a correction term, which will depend on a framing of the 3-manifold M . The
result is that the BV cohomology class of the “fixed” partition function is independent of
the gauge-fixing Lagrangian but depends on the framing.

73
3.1 BV Chern-Simons theory

3.1.1 Setup

Let M be a compact oriented 3-manifold and G a simple and simply connected Lie group
with Lie algebra g. Let P → M be a principle G-bundle. By Lemma 1.2.7, there exists a
section s : M → P . The space of fields is FM = Conn(P ) which we identify with Ω1 (M, g)
using the section s. Fixing an invariant symmetric bilinear form on g satisfying assumption
1.2.10, we define the Chern-Simons action functional
Z
1 1
SCS [A] = hA, dAi + hA, [A, A]i
M 2 6
and we have that its exponential
 
i 1
exp SCS , ~=
~ 2πk
1
where ~ = 2πk
for some k ∈ Z, is independent of the choice of section s and invariant under
gauge transformations

A 7→ Ag = g −1 Ag + g −1 dg, g ∈ C ∞ (M, G).

3.1.2 BRST operator

The Lie algebra of the gauge group C ∞ (M, G) is C ∞ (M, g) with infinitesimal action ρ#
A (c) =
dA c. The space of minimal BRST fields is1

min
FBRST = Ω1 (M, g) ⊕ ΠΩ0 (M, g) 3 (A, c)

with BRST operator

QA = dA c (3.1)
1
Qc = [c, c]. (3.2)
2
(3.3)
1
Since we have trivialized P , we can identify the Ad P -valued forms with Ω• (M, g).

74
3.1.3 BV formulation

The minimal space of BV fields as in Equation (2.81) is then

F = ΠT ∗ FBRST
min
= ΠΩ0 (M, g) ⊕ Ω1 (M, g) ⊕ ΠΩ2 (M, g) ⊕ Ω3 (M, g) 3 (c, A, A+ , c+ ) (3.4)

where we identify Ωk (M, g) with (Ωn−k (M, g))∗ using the pairing
Z
ωBV (A, B) = hA, Bi (3.5)
M

which, extended to F, becomes an odd symplectic form. Notice that we can write the BV
space of fields as
F = ΠΩ• (M, g), (3.6)

where the Π simply means that we are shifting the “natural” parity of the superspace
Ω• (M, g) induced by form degree by one. The BV action, as given in Equation (2.82),
is Z Z
α 1
SBV = SCS + Q(φ )φ+
α =S+ +
hA , dA ci + hc+ , [c, c]i (3.7)
M 2 M
The following is a crucial observation. If we define the “superfield”

A = c + A + A+ + c+ ∈ Ω• (M, g), (3.8)

the Chern-Simons action can be rewritten in the following intriguing fashion:

Z Z
+ + 1 1
SBV [c, A, A , c ] = SCS [A] = hA, dAi + hA, [A, A]i . (3.9)
M 2 M 6

To see this, notice that in terms of “total parity” (i.e. form parity plus ghost parity) all
fields are odd. Hence both terms in the action are totally symmetric in all fields. Since only
the terms of total form degree 3 contribute to the integral, we get
Z Z Z
1 1 1
hA, dAi + hA, [A, A]i = hA, dAi + hA+ , dci + hc, dA+ i
M 2 M 6 2 M
Z
1 1
+ hA, [A, A]i + hc+ , [c, c]i + hA+ , [A, c]i
M 6 Z
2
Z
+ 1
= SCS [A] + hA , dA ci + hc+ , [c, c]i
M 2 M

using that dA c = dc + [A, c].

75
3.1.4 The Quantum Master Equation

Let us now argue that the BV-extended Chern-Simons action formally satisfies the QME
i
∆e ~ SBV = 0. (3.10)

To this end, we first show that the action satisfies the classical master equation (SBV , SBV ) =
0. The simplest way to see this is to use the superfield formalism and notice that
δ
Q = (S, ·) = FA .
δA
Now
Z
1
Q(S) = hA, dFA i + hFA , [A, A]i
2
ZM
1 1 1
= hA, d[A, A]i + hdA, [A, A]i + h[A, A], [A, A]i.
M 2 2 2
The first two terms combine into a total derivative, and the last term vanishes by the Jacobi
identity and invariance of the pairing. For the BV Laplacian it is better to use field-antifield
notation.2
XZ δ δ δ δ
∆= + i i
+
i M δ(A ) δA δ(c ) δci
+ i

where we have expanded the fields in an orthonormal basis Ti . The only contribution to
∆(SBV ) comes from the terms hA+ , [A, c]i and hc+ , [c, c]i and is proportional to
XZ
∆(S) ∝ cj fjii (δ (3) (x, y))2 .
i (x,y)∈M ×M
P
Since we do not want to deal with the square of this delta form, we assume that i fjii = 0
for all j. This is condition is equivalent to unimodularity:

Definition 3.1.1. A Lie algebra g is called unimodular if tr adx = 0 for all x ∈ g.

We conclude that if g is unimodular, the Quantum Master Equation is satisfied at least

formally, since  
i
S 1 i
∆e ~ = (S, S) − i~∆(S) e ~ S .
2
2
The functional derivative is defined by the requirement δ j
δφi (x) φ (y) = δij δ (3) (x, y), where δ (3) (x, y) is the
integral kernel of the identity map on Ω• (M ).

76
It is a fact that simple groups, which we are restricting ourselves to, always have unimodular
Lie algebras, see [Kna96, Corollary 8.31].3

3.2 Gauge fixing using background fields

We know that in finite dimensions the asymptotics of the path integral depend only on
the neighourhood of the critical points. Bearing this in mind we fix a critical point of the
Chern-Simons action functional – i.e. a flat connection – and gauge fix the theory in the
neighbourhood of that flat connection to compute the contribution to the partition function
from that critical point. It is a subtle and actually open question how to piece together the
contributions from the different critical points in this infinite-dimensional context.4

3.2.1 Fixing a background field

As discussed in Section 1.3, the critical points of the Chern-Simons action functional are
the flat connections on P . For the following, we fix a flat connection A0 ∈ Ω1 (M, g), i.e.
dA0 + 21 [A0 , A0 ] = 0, and work in terms of a fluctuation  defined by A = A0 + Â. Expressing
the action in terms of this decomposition, we obtain, using that the curvature of A0 vanishes,
Z
1 1 A0
SCS [A0 + Â] = SCS [A0 ] + hÂ, dA0 Âi + hÂ, [Â, Â]i =: SCS [A0 ] + SCS [Â]. (3.11)
M 2 6
The first term is simply the Chern-Simons action evaluated at the flat connection A0 , the sec-
ond term is the Chern-Simons action of Â, but with differential twisted by the flat connection
A0
A0 . We denote this action by SCS .

3.2.2 The gauge fixing Lagrangian

In the spirit of the perturbation theory explained in the last Section, we now gauge fix the
A0
quadratic part of SCS and treat the cubic part as a perturbation. The gauge-fixing in the
3
In loc. cit. it is shown that semisimple (and in particular simple) Lie groups are unimodular, i.e. that
their left and right Haar measures coincide. This is equivalent to the fact that | det Ad g| = 1 for all g ∈ G,
which implies unimodularity of the Lie algebra.
4
See also the upcoming publication [MW21].

77
BV formalism consists of two steps: First, finding a decomposition of the BV space of fields
F = Y × Y 0 such that there exists a gauge-fixing Lagrangian L ⊂ Y 0 , and second, choosing
A0
such a gauge-fixing Lagrangian. The quadratic part of SCS is
Z
A0 1
Sf ree := hÂ, dA0 Âi. (3.12)
2 M
One possibility to gauge fix it is given by Hodge theory. Below we recall briefly how this
works.

Hodge decomposition for dA0

Let g be a Riemannian metric on M . The metric induces a Hodge star

∗ : Ωk (M, g) → Ω3−k (M, g)

and a Hodge pairing on forms Z

(ω, τ ) = hω, ∗τ i (3.13)
M

which turns Ω• (M, g) into a pre-Hilbert space. With respect to the Hodge pairing, the
operator dA0 has a formal adjoint d∗A0 uniquely defined by the property

(dA0 ω, τ ) = (ω, d∗A0 τ ). (3.14)

Central for Hodge theory is the (twisted) Hodge-deRham Laplacian

∆A0 : = d∗A0 dA0 + dA0 d∗A0 : Ω• (M, g) → Ω• (M, g) (3.15)

Forms in the kernel of ∆A0 are called harmonic forms and denoted Harm•A0 (M ).

Lemma 3.2.1. ω is harmonic if and only if it is both closed and co-closed: dA0 ω = d∗A0 ω = 0.

Proof. A closed and co-closed form is obviously harmonic. For the opposite, consider

0 = (ω, ∆A0 ω) = (d∗A0 ω, d∗A0 ω) + (dA0 ω, dA0 ω).

The terms on the right vanish if and only if ω is both closed and co-closed.

78
Theorem 3.2.2 (Hodge theorem). The map

Harm•A0 → HA• 0 (M, g) (3.16)

ω 7→ [ω] (3.17)

is an isomorphism.

For a proof see e.g. [Nic07]. The key to gauge fixing is the Hodge decomposition, which
we now state. We denote the spaces of dA0 -exact and -coexact forms by dA0 Ω• (M, g) =:
Ω•ex (M, g) and d∗A0 Ω• (M, g) =: Ω•coex (M, g) respectively.

Theorem 3.2.3. We have

Ω• (M, g) = Harm•A0 (M, g) ⊕ Ω•ex (M, g) ⊕ Ω•coex (M, g) (3.18)

where all sums are orthogonal with respect to (·, ·).

A proof can be found in the same reference as above.

Lorenz gauge as gauge-fixing Lagrangian

Using the Hodge decomposition we can write

F = Y × Y0 (3.19)

with

Y = Harm•A0 (M )
Y 0 = Ω•ex (M, g) ⊕ Ω•coex (M, g)

Since the direct sum in (3.18) is orthogonal, the decomposition (3.19) is symplectic.

Definition 3.2.4. The Lorenz gauge Lagrangian is the space of coexact forms,

Lg := Ω•coex (M, g) = im d∗A0 = ker d∗A0 Y0

. (3.20)

Notice that in Equation (3.20) we have im d∗A0 = ker d∗A0 since we are only considering forms
in the orthogonal complement of harmonic forms. The next Lemma tell us this is indeed a
Lagrangian.

79
Lemma 3.2.5. Lg ⊂ dA0 Ω• (M, g) ⊕ d∗A0 Ω(M, g) is Lagrangian, i.e. both isotropic and
coisotropic.

Proof. Isotropy (L ⊂ L⊥ ) follows immediately from (d∗A0 )2 = 0 and integration by parts:

Z Z
∗ ∗ ∗ ∗
ωBV (dA0 ω, dA0 τ ) = dA0 ω ∧ dA0 τ = ω ∧ (d∗A0 )2 τ = 0
M M

shows isotropy. To see that it is also coisotropic (L⊥ ⊂ L) take a form 0 6= α ∈ L. Then

0 6= (dA0 α, dA0 α) = ±ωBV (dA0 α, d∗A0 (∗α)).

/ L⊥ . Since all non-zero forms in Ω•ex (M, g) are of the form dA0 α, this shows
Hence dA0 α ∈
that Ω• (M, g) ∩ L⊥ = {0}.

3.2.3 Inverting the quadratic term

R
Notice that the quadratic term M
hÂ, dA0 Âi has a unique critical point on Lg , namely  = 0.
The operator dA0 : Ωkcoex (M, g) → Ωk+1
ex (M, g) is an isomorphism, and can be inverted as
follows. Let ω = d∗A0 τ ∈ Ωkcoex (M, g). Then dA0 ω = dA0 d∗A0 τ and, since (d∗A0 )2 = 0, we have

d∗A0 dA0 ω = d∗A0 dA0 d∗A0 τ = (d∗A0 dA0 + dA0 d∗A0 )d∗A0 τ = ∆A0 ω.

But on the orthogonal complement of harmonic forms, the Laplacian is invertible, so we can
define
KA0 := d∗A0 ◦ ∆−1 k+1 k
A0 : Ωex (M, g) → Ωcoex (M, g). (3.21)

It then follows that dA0 KA0 ω = KA0 dA0 ω = ω for all ω ∈ Lg . Hence KA0 is an inverse to dA0
on Lg . See also Figure 3.1 above.
Remark 3.2.6. We can extend KA0 to an operator Ω• (M, g) → Ω•−1 (M, g) by defining

KA0 = d∗A0 ◦ (∆A0 + PA0 )−1 , (3.22)

where PA0 is the orthogonal projection to harmonic forms. In terms of the Hodge decom-
position this just means extending the operator KA0 by 0 on coexact and harmonic forms.
This operator satisfies
dA0 KA0 + KA0 dA0 = id − PA0 (3.23)

80
 Harmk+1 ⊕ Ωk+1
ex ⊕
d∗ Ωk+1
coex

Harmk ⊕ Ωkex ⊕ d Ωkcoex

Figure 3.1: Hodge decomposition at degree k (subscripts and arguments suppressed). d,d∗
are both isomorphisms restricted to the spaces but not inverse to each other, rather, their
composition is the Laplacian ∆.

This means that KA0 is a parametrix for dA0 (an inverse up to smoothing operators). In
the language of homological algebra, the map KA0 defines a chain homotopy between the
identity map and the projection to harmonic forms. The triple (ιA0 , KA0 , PA0 ), where ιA0 is
inclusion of harmonic forms, is called a contracting triple for the complex (Ω• (M, g), dA0 ).
From the definition, it is clear that KA0 satisfies KA2 0 = 0 and KA0 ◦ ιA0 = 0.5 For more on
contracting triples see e.g. [CM08].

3.2.4 The propagator

The propagator, loosely speaking, is the integral kernel of the map KA0 . There are different
conventions for integral kernels. To reflect the topological nature of Chern-Simons theory,
we use here the “topological kernel” of the map KA0 (this is the terminology of de Rham
[Rha84]). This is a de Rham 2-current η̂ 6 on M × M with values in g × g such that7
Z
(KA0 ω)x = hη̂(x,y) ∧ ωy i23 (3.24)
y∈M

where we write pairing between currents and forms as integrals, as customary also for dis-
tributions, and we define

hα ⊗ ξ1 ⊗ ξ2 ⊗ ξ3 i23 = hξ2 , ξ3 iα ⊗ ξ1 .
5
These last two equations mean the maps (ιA0 , KA0 , PA0 ) between the complexes HarmA0 and Ω• (M, g)
are a special deformation retract, see e.g. [Cra04].
6
Roughly speaking, de Rham currents are to forms what distributions are to functions. For more back-
ground the reader is referred to the excellent original text [Rha84].
7
R
if we were instead to use the “metric kernel”, it would be M η̂ ∧ ∗y instead

81
From the fact that ∆A0 is an elliptic operator, it follows8 that η̂ can be represented by a
smooth 2-form η ∈ Ω2 (M × M − diag, g ⊗ g). One of the main results of Axelrod and Singer
[AS91],[AS94] is that the singularities of η are tame enough to ensure that η extends to the
differential-geometric blow-up of the diagonal Bldiag (M × M ), which, roughly speaking, is
defined by replacing the diagonal with its unit sphere bundle ST diag. For reasons that will
become clear below, we denote

C2 (M ) := Bldiag (M × M ) (3.25)

Proposition 3.2.7. C2 (M ) has the following properties.

• C2 (M ) is a smooth manifold with boundary.

• The boundary of C2 (M ) is diffeomorphic to the unit sphere bundle over M : ∂C2 (M ) ∼

=
ST M → M .

• C2 (M )◦ = C2 (M ) − ∂C2 (M ) = M × M − diag.

• C2 (M ) is compact (if M is, which we always assume).

A proof can be found in [Sin03] or the papers of Axelrod and Singer cited above.

Theorem 3.2.8 (Axelrod-Singer [AS91]). There is a smooth 2-form η ext ∈ Ω2 (C2 (M ), g⊗g)
such that ι∗M ×M −diag η ext = η.

The smooth 2-form η ext will be the propagator that we work with. From now on we will
drop the superscript ext. It has the following important properties.

Proposition 3.2.9. The propagator η ∈ Ω2 (C2 (M ), g × g) satisfies:

1. Denote χi an orthonormal basis of harmonic forms and χi = ∗χi . Then

X
dA0 η = ±π1∗ χi π2∗ χi . (3.26)
i

Here π1 , π2 : C2 (M ) → M are the extensions to C2 (M ) of the restrictions of π1 , π2 : M ×

M → M to M × M − diag.
8
See loc. cit.

82
 2. Denote ι∂ : ∂C2 (M ) ,→ C2 (M ) the inclusion. Then ι∗∂ η is a global angular form on the
sphere bundle ∂C2 (M ) ∼= ST M .

3. Denote T the extension of the map (x1 , x2 ) 7→ (x2 , x1 ) to C2 (M ). Then

T ∗ η = −η (3.27)

4. For all i we have Z

hη(x, y) ∧ χi i23 = 0. (3.28)
y∈M

5. We have Z
hη(x, y) ∧ η(y, z)i23 = 0. (3.29)
y∈M

Proof. Equation (3.26) is a direct consequence of the parametrix property Equation (3.23)
(notice the right hand side is an integral kernel for the projection to harmonic forms).
Equation (3.28) follows from KA0 ◦ ιA0 = 0, and Equation (3.29) follows from KA2 0 = 0. For
the remaining two points we refer to [CM08, Section 4].

3.3 The effective action

In the presence of zero modes we can not integrate over all fields at once, however, we can
define an effective action on zero modes as a formal BV pushforward. To this end we split
the fluctuation  = α + a, where a ∈ Harm•A0 (M, g). The quadratic part of the action then
becomes
hα + a, dA0 (α + a)i = hα, dA0 αi = (α, ∗dA0 α). (3.30)

For α ∈ Lg the operator ∗dA0 becomes invertible. The cubic term becomes hα + a, [α +
a, α + a]i. Expanding this term we obtain 3-valent vertices, but any number of half-edges
issuing from these vertices can end in a zero mode a field (see below for the precise Feynman
rules). We now use the superspace formulation of the perturbative integral: We will get a
superdeterminant from the quadratic part and a signature factor from the even part of the

83
quadratic part. All in all, we obtain
Z
A0 i 1/2
ZCS (a; g) = e ~ SCS [A0 ] µBV
L
!
iπ
i
−1/2 4
sign ∗dA0
1/2
X (−i~)χ(Γ)
:= |e ~ S{z
CS [A0 ]
} |sdet ∗dA0 L
e Leven
µY F (Γ) . (3.31)
|Aut(Γ)|
|Γ
=:Z
{z }
back
=:Zf ree {z }
=:Zpert

Here Zback , Zf ree are traditionally called the 0- and 1-loop part, respectively, but in this
context this terminology is misleading because in the presence of zero modes the perturbation
series Zpert contains 0- and 1-loop graphs (which are absent if there are no zero modes).

Definition 3.3.1. The effective action on zero modes with background A0 is defined by

A0 i
ZCS (a; g) = Zback Zf ree e ~ Seff (a) (3.32)

A standard combinatorial exercise shows

Proposition 3.3.2. The effective action is given by

X (−i~)(loops(Γ))
Seff (a) = F (Γ) (3.33)
Γ connected
|Aut(Γ)|

(of course, this uses the fact that F (Γ1 t Γ2 ) = F (Γ1 )F (Γ2 ), which will become clear below).

3.3.1 Feynman graphs and rules

In Equation (3.31) the sum is over all trivalent graphs with leaves, i.e. there are three
half-edges emanating from every vertex that can either be connected to another half-edge
or end in a leaf. We do not allow tadpoles9 (edges connecting a vertex to itself, also
known as short loops). Formally, these graphs are given by a set by a quadruple Γ =
(H(Γ), V (Γ), L(Γ), E(Γ)) where

• H(Γ) is a finite set (the half-edges),

9
The unimodularity condition (Definition 3.1.1) implies that any graph containing a tadpole evaluates to
zero under the Feynman rules, so we might as well disregard them from the start

84
 • V (Γ) is a partition of H(Γ) into sets of cardinality three,

• L(Γ) is any subset of H(Γ) (the leaves)

• E(Γ) is a perfect matching of H(Γ) − L(Γ) (the edges).

The set of all such graphs is denoted Gr. We think of the leaves as decorated by an a field.
Let us define first the graph configuration space:

Definition 3.3.3. For every graph Γ, the open graph configuration space is

CΓ◦ (M ) := {c : V (Γ) → M | c(v) 6= c(w) if v 6= w} (3.34)

If c ∈ CΓ◦ , we can extend c to H(Γ) (the set of half-edges) by defining c(h) = c(v) if h ∈ v.
This extension is well-defined since vertices partition the set of half-edges.
To formulate the Feynman rules, let Ta be an orthonormal basis of the Lie algebra and define
fabc = hTa , [Tb , Tc ]i. Expand η = η ab Ta ⊗ Tb and a = ai Ti .

Definition 3.3.4 (Feynman rules). A labeling of a graph γ is a map H(Γ) → {1, . . . , dim g}.
The graph differential form is a differential form ωΓ ∈ Ω• (CΓ◦ ) defined by
X Y Y Y
(ωΓ )f := fl(h1 )l(h2 )l(h3 ) η l(h1 )l(h2 ) (c(h1 ), c(h2 )) al(h) (c(h)). (3.35)
l v=h1 ,h2 ,h3 ∈V (Γ) e=h1 ,h2 h∈L(Γ)

We then define the Feynman rules map

F : Gr → R (3.36)

by Z
F (Γ) = ωΓ (3.37)
CΓ◦ (M )

For convenience, we also the define the normalized versions

(−i~)χ(Γ)
ω̃Γ = ω(Γ)
|Aut(Γ)|
(−i~)χ(Γ)
F̃ (Γ) = F (Γ)
|Aut(Γ)|

85
 a a

(a) Γ1 (b) Γ2

Figure 3.2: Two graphs in the effective action

Example 3.3.5. Let Γ1 be the theta graph of Figure 3.2a. Then we have
X
ωΓ1 (x, y) = fijk flmn η ik (x, y)η jl (x, y)η kn (x, y)
i,j,k,l,m,n

and
(−i~)2
ω̃Γ1 (x, y) = ωΓ1 (x, y).
12
On the other hand, for the graph of Figure 3.2b with residual fields, we have
X
ωΓ2 = fijk flmn ai (x)η jm (x, y)η kn (x, y)al (y)
i,j,k,l,m,n

and
−i~
ω̃Γ2 = ωΓ2 .
4

In principle, F (Γ) does not need to be finite, as the open configuration space CΓ◦ is not
compact. However, this is not the case. Again, this was first proven in generality by Axelrod
and Singer.

Theorem 3.3.6 ([AS91],[AS94]). F (Γ) is finite for all graphs Γ.

The proof goes through the construction of a compactification CΓ (M ) of CΓ◦ (M ) such that
the form ωΓ extends to CΓ (M ). This is a generalization of the compactification C2 (M )
discussed in Section 3.2.4 above. Instead of giving an explicit construction, we simply state
the properties of this compactification that are important for us. Let us fix some notation
first.

Definition 3.3.7 (Configuration space of a finite set). For a finite set S denote CS (M )◦
the open configuration space given by maps S ,→ M . For convenience denote C[n] (M )◦ =:
Cn (M )◦ .

86
Definition 3.3.8 (Reduced configuration space). If V is a vector space, the group V o R>0
acts on CS (V ) by scaling and translations. C̃S (V )◦ denotes the quotient of CS (V ) under this
group action, and C̃S (V ) the corresponding compactification.

Proposition 3.3.9. There is a compact smooth 3n-dimensional manifold with corners C[n] (M ) :=
Cn (M ) that satisfies:

1. There is a stratification
[
Cn (M ) = Cn (M )(k)

with Cn (M )(k) compact, Cn (M )(k−1) ⊂ Cn (M )(k) , such that Cn (M )(k) \ Cn (M )(k−1) is

a smooth manifold of dimension 3n − k and

Cn (M )(0) \ Cn (M )(1) = Cn◦ (M ).

2. The maps πij : Cn◦ (M ) → C2 (M )◦ given by (x1 , . . . , xn ) 7→ (xi , xj ) admit smooth exten-
sions πij : Cn (M ) → C2 (M ) to the respective compactifications.

3. The boundary of Cn (M ) is ∂Cn (M ) = Cn (M )(1) is given by

G
∂Cn (M ) = ∂S Cn (M ),
S⊂[n],|S|≥2

where ∂S Cn (M ) is a fiber bundle over C([n]\S)t{∗} (M ). The fiber over a configuration ι

is the reduced configuration space C̃S (Tι (∗)M ) of |S| points in Tι(∗) M .

The second point in this proposition implies that there is a form ωΓext on CΓ (given by pulling
back extended propagators) such that ωΓext CΓ◦
= ωΓ . Hence
Z Z
ωΓ = ωΓext
CΓ◦ (M ) CΓ (M )

but the integral on the right hand side is of a smooth differential form over a compact
manifold, so it is finite.

87
3.3.2 The superdeterminant and the Ray-Singer torsion

It is a surprising fact that the (regularized) superdeterminant of the quadratic part of the
theory can be explicitly computed and is given by a known invariant, the so-called Ray-
Singer torsion [RS71]. This was first shown by Schwarz in 1978 [Sch78] (this was maybe the
first explicit connection between partition functions of topological theories and topological
invariants). The aim is to compute the regularized superdeterminant
     −1
sdet ∗dA0 Lg = det ∗dA0 Leven det ∗dA0 L
odd

Here the determinants are to be understood as zeta-regularized determinants: The idea is to

define for an operator A the determinant by
X
log det A = lim (log λ)λ−s .
s→0
λ∈spec(A)\{0}

For elliptic operators, the sum on the right hand side always converges for s > 1 and has
an analytic extension to 0. The limit is understood in this sense, see e.g. [BGV03]. Denote
(i) (i)
by ∆A0 the restriction of the twisted Hodge-de Rham Laplacian to i-forms, and det ∆A0 its
zeta-regularized determinant. Then we have the following definition:

Definition 3.3.10 (Ray-Singer torsion [RS71]). Let µH denote the density on the space
of dA0 -harmonic forms corresponding to any orthonormal basis. The Ray-Singer torsion or
analytic torsion is
3
Y (i) (−1)i i
τ (M, A0 ) = det(∆A0 )− 2 µH . (3.38)
i=0

The result is that the superdeterminant in question is related to the inverse of the Ray-Singer
torsion:
 
Lemma 3.3.11. For the zeta-regularized superdeterminant sdet ∗dA0 Lg
we have

−1/2 −1/2
sdet ∗dA0 L
= τ (M, A0 )1/2 µH . (3.39)

Notice that the right hand side of the Equation (3.39) above is a number, the product of
regularized determinants.

88
  1/2
2
Proof. Note that on Lg we have (∗dA0 ) = ∆A0 , thus det ∗dA0 odd/even
Ωcoex
= det ∆A0 odd/even
Ωcoex
and we can rewrite the superdeterminant as
   1/2  −1/2
sdet ∗dA0 Lg = det ∆A0 Ω1coex
det ∆A0 Ωeven
.
coex

Consider the Hodge decomposition of the complement of harmonic forms:

Ω3ex d∗

Ω2ex d d∗
Ω2coex
∗ (3.40)
d
Ω1ex Ω1coex
d∗

d
Ω0coex

(subscripts and arguments of d and Ω are suppressed to maintain a minimum of readabil-

ity). All arrows in this diagram are isomorphisms, and commute with the Hodge-de Rham
Laplacian. The right hand column is precisely the gauge-fixing Lagrangian Lg = Ω•coex . The
fact that all isomorphisms commute with the Laplacian means that there essentially only
two “independent” pieces in the Laplacian: The Laplacian restricted to 0-forms and the
Laplacian restricted to coexact one-forms. More precisely, we have

det ∆ Ω0
= det ∆ Ω1ex
= det ∆ Ω2coex
= det ∆ Ω3
(3.41)

and
det ∆ Ω1coex
= det ∆ Ω2ex
. (3.42)

We can now combine this with the equation

det ∆(i) = det ∆ Ωicoex

det ∆ Ωiex
(3.43)

to obtain

 1/2  −1/2
sdet ∗dA0 L
= det ∆ det ∆ Ωeven
Ω1coex coex
 −1/2  −1/2
(1) 1/2 (0) −1/2


= (det ∆ ) det ∆ Ω1 det ∆ det ∆ Ω2coex
ex

(det ∆(1) )1/2

=
(det ∆(3) )3/2

89
Now we use the fact det ∆(1) = det ∆(2) (which again follows from the fact that the Hodge
star intertwines the two Laplacians) to rewrite this as
3
(det ∆(1) )1/2 1 det ∆(2) Y
(i)
(−1)i i
= = det(∆ ) 2 .
(det ∆(3) )3/2 (det ∆)1/2 (det ∆(3) )3/2 i=0

The right hand side is, up to the density factor, exactly the inverse of the Ray-Singer torsion
[RS71]
3
Y (i) (−1)i i
τ (M, A0 )µ−1
H = det(∆A0 )− 2 .
i=0

Hence we conclude that

−1/2 −1/2
sdet ∗dA0 L
= τ (M, A0 )1/2 µH .

By the Cheeger-Müller Theorem ([Che77],[Mül78],[Che79]) the Ray-Singer torsion is equal

to the Reidemeister [Rei35] torsion (see e.g. [Nic03] for background on the Reidemeister
torsion) of the corresponding representation of π1 (M ), which is a topological invariant of M ,
i.e. independent of the metric g. Notice that in the non-acyclic case neither the product of
determinants nor the density µH are independent of g, only their product is. However, the
combination appearing in our definition of the path integral is precisely

−1/2 1/2
sdet ∗dA0 L
µH == τ (M, A0 )1/2 ,

which is independent of g.

3.3.3 The phase factor

Similar to the determinant, we can also zeta-regularize the signature of ∗dA0 Ω1coex
:
X
sign ∗dA0 := lim sign λ|λ−s | (3.44)
s→0
λ6=0

(again this limit is understood in the sense of analytic continuation). Also this invariant can
be expressed in terms of known quantities. To this end we need the following Lemma.10
10
This lemma and its proof were explained to the author by [Link].

90
Lemma 3.3.12. The spectrum of dA0 ∗ : Ωodd odd
ex → Ωex is symmetric around 0.

Proof. Let f ∈ Ω0 (M, g) be an eigenfunction of the Laplacian, ∆A0 f = λ2 f . Then we claim

that
ω±λ = dA0 ∗ dA0 f ± λdA0 f (3.45)

is an eigenform of ∗dA0 Ωodd

of eigenvalue ±λ. Indeed, we have
ex

dA0 ∗ (±λdA0 f + dA0 ∗ dA0 f ) = ±λdA0 ∗ dA0 f + dA0 ∆A0 f

= ±λdA0 ∗ dA0 f + dA0 λ2 f = ±λ(±λdA0 f + dA0 ∗ dA0 f ).

We claim that all eigenforms of ∗dA0 are of the form ω±λ as defined in Equation (3.45) above.
Indeed, if ω ∈ Ω1ex ⊕ Ω3ex is an eigenform of ∗dA0 of eigenvalue λ, then - since ∗dA0 squares
to ∆ on this space - it is an eigenform of ∆A0 of eigenvalue λ2 . But ∆A0 maps Ω1 and Ω3 to
itself, hence the 1- and 3-form components of ω must be eigenforms of ∆A0 of eigenvalue λ2 .
But these eigenforms are precisely given by dA0 f – resp. dA0 ∗ dA0 f – for f and eigenfunction
of ∆(0) of eigenvalue λ2 . Hence ω is equal to ω±λ (for one of two signs). We conclude that
the spectrum is given precisely by {±λ|λ ∈ spec(∆(0) )}.

As a direct corollary we have:

Corollary 3.3.13.

sign ∗dA0 Ω1 coex

= sign ∗dA0 + dA0 ∗ Ωodd
=: ψ(A0 , g) (3.46)

Here ψ(A0 , g) is the Atiyah-Patodi-Singer eta invariant [APS73],[APS75] of the Dirac oper-
ator L− := ∗dA0 + dA0 ∗ : Ωodd → Ωodd . In contrast to the torsion, it is not invariant under
change of the metric g. This is the first sign of the anomalous behavior of the perturbative
quantum Chern-Simons theory. We will return to this issue in section 3.5 below.

3.4 The Quantum Master Equation, again

In Section 3.1.4 we have argued that the Quantum Master Equation formally holds since we
have assumed our Lie group to be simple and thus unimodular. We thus expect that the

91
conclusion of Theorem 2.4.7 holds, i.e. that the effective action also satisfies the Quantum
Master Equation. However, since the theorem does not apply to the infinite-dimensional
setting, the Quantum Master Equation has to be proven “by hand”.

3.4.1 The Laplacian on the space of residual fields

The space of residual fields is the space of harmonic forms,

Y = ΠHarm•A0 (M, g) ∼
= ΠHA• 0 (M, g).

It carries the symplectic form Z

ωY (a, b) = ha, bi.
M
To express the Laplacian it is helpful to express the symplectic form in Darboux coordinates.
To this end, choose orthonormal bases (w.r.t. the Hodge pairing (·, ·)) χi(k) of HAk 0 (M, g) for
k = 0, 1 and we let χi(k) = ∗χi(3−k) for k = 2, 3. We also choose an orthonormal basis Ta of
(k)
g and expand the the forms accordingly χi(k) = χi,a
(k) Ta . Denote by zi,a the coordinates with
respect to this basis. Then the symplectic form is expressed as
X (3) (0) (2) (1)
ωBV = dzi,a ∧ dzi,a + dzi,a ∧ dzi,a
i,a

and the BV Laplacian can be expressed as

X ∂ ∂ X ∂ ∂
∆Y = (3) (0)
+ (2) (1)
(3.47)
i,a ∂zi,a ∂zi,a i,a ∂zi,a ∂zi,a

The following Lemma is crucial to the proof of the QME.

Lemma 3.4.1. Denoting the propagator by η ∈ Ω2 (C2 (M ), g ⊗ g) we have

dA0 η = ∆Y π1∗ aπ2∗ a. (3.48)

(k)
Proof. First expand the left hand side in the basis χi,a to obtain (cf. Proposition 3.2.9)
3 X
X (k) (3−k)
dA 0 η = π1∗ (χi,a )π2∗ (χi,a ).
k=0 i,a

P i,a(k)
Expanding a = i,a,k z(k) χi,a , the Lemma follows immediately from Equation (3.47).

92
Pictorially this Lemma is expressed as dA0 ( ) = ∆Y ( a a )= χ χ , i.e. the

differential “cuts” the edge corresponding to a propagator (in the last picture summation
over indices of χ is implied).

3.4.2 Proof of Quantum Master Equation

The main theorem of this section is that the effective action satisfies the Quantum Master
Equation:

Theorem 3.4.2. We have

i
∆Zpert = ∆Y e ~ Seff = 0. (3.49)

A0
Since Zback , Zf ree do not depend on a this implies that ∆Y ZCS (a) = 0. There are two main
steps in the proof. The first one is to rewrite ∆Z as a sum of integrals over boundary faces
of the compactified configuration space. The second is to show that all such integrals vanish.
This is indeed a general cooking recipe for proofs of such equations in theories whose propa-
gators admit extensions to such compactifactions, starting in [AS91],[AS94],[BT94],[Kon94].
See also [Cat+95] for a more pedestrian introduction.

Applying Stokes’ theorem

Proposition 3.4.3. Using the notation of Definition 3.3.4, write

XZ
Zpert = ω̃Γ .
Γ CΓ (M )

Then
XZ
∆Y Zpert = dA0 ω̃Γ . (3.50)
Γ CΓ (M )

Proof. ω̃Γ is a product of propagators η and zero modes a. From the Leibniz rule and the
fact that dA0 a = 0, we have that
X
dA 0 ω = ω̃Γe
e∈E(Γ)

93
where ω̃Γe denotes the form where to the edge e we associate the form dA0 η = π1∗ χi,a ∗ i,a
P
i,a,k (k) π2 χ(k) .
Thus, we can express
XZ XZ (−i~)χ(Γ) e
dA0 ω̃Γ = ω
Γ CΓ (M ) Γem CΓ |Autm (Γ)| Γ

as a sum over graphs with one marked edge. Notice that here we have replaced Aut(Γ) by
automorphisms Autm (Γ) of marked graphs to account for the fact that marking different
edges might lead to automorphic marked graphs. On the other hand, we have
XZ X X Z
∆ ω̃Γ = ωΓl1 ,l2
Γ CΓ (M ) Γ l1 6=l2 ∈L(Γ) CΓ

where ωΓl1 ,l2 denotes the form that results from the usual Feynman rules with ∆ applied to
the residual fields at the two leaves l1 , l2 . Notice that if l1 , l2 are placed at the same vertex,
ωΓl1 ,l2 contains a term of the form fjii = 0. Hence, we can rewrite ∆Z by
XZ (−i~)χ(Γ) l1 ,l2
∆Zpert = ω
l ,l2 CΓ |Autm (Γ)| Γ
1
Γm

where the sum goes over graphs with a pair of marked leafs. Now, by Lemma 3.4.1 we have
ωΓe = ωΓl1 ,l2 if l1 , l2 are placed at the same vertices as the start and end of e. The claim
now follows from noticing that there is an obvious bijection between automorphism classes
of graphs with a marked edge and automorphism classes with a pair of marked vertices by
simply connecting the two marked leaves (with inverse given by cutting the marked edge).

We thus conclude that, by Stokes’ theorem,

XZ Z
∆Y Zpert = dA0 ω̃Γ = ω̃Γ . (3.51)
Γ CΓ (M ) ∂CΓ

Vanishing of the boundary contributions

As explained in Proposition 3.3.9, the boundary of the compactified configuration space has
different components, one for every subset of V (Γ) of cardinality at least 2:
G
∂CΓ = ∂S Γ.
S⊂V (Γ),|S|≥2

94
We will make extended use of the fact that for a fiber bundle F ,→ E  M , we have a
generalized Fubini theorem for integration along the fiber
Z Z Z 
ω= ω .
E M F

We will apply this to ∂S CΓ , which is a fiber bundle over CΓ/S with fiber over c given by
C̃S (Tc([S]) M ).
One has to distinguish two cases, the case where |S| = 2 (the so-called “principal faces”)
and the case |S| ≥ 3 (the so-called “hidden faces”).

Lemma 3.4.4. We have Z

X X
ωΓ = 0. (3.52)
Γ S⊂V (Γ),|S|=2 ∂S CΓ

Let us first consider the case |S| = 2. Let S = {v1 , v2 }. The boundary stratum is a sphere
bundle over the “small” diagonal v1 = v2 . If there is no edge connecting v1 , v2 , then the
form ωΓ is regular on this diagonal and hence has no form degree on the fibers of this
bundle. Hence the corresponding contribution vanishes. If there is an edge e = S, then the
corresponding boundary face is a sphere bundle over CΓ/e (here Γ/e denotes Γ with the edge
e contracted). By normalization of the propagator corresponding to e, integrating over the
fiber of this sphere bundle yields ±1 and we have
Z Z
ωΓ = ωΓ/e .
∂e C Γ CΓ/e

These contributions do not vanish individually, but only after we sum over all graphs. The
graphs Γ/e contain a single 4-valent vertex, and there are three different possibilities for this
4-valent vertex to arise, see figure 3.3. Summing over these three possibilities we obtain the
P
4-vertex weight vabcd = e faeb fced + faec fbed + faed fbec which vanishes by the Jacobi identity
P
e f[abe fc]ed = 0 (the square brackets denote antisymmetrization over uncontracted indices).
Thus, after summing over all graphs, these contributions cancel out.11
In contrast to the principal faces, where we have to sum over all graphs to ensure vanishing,
the contributions from hidden faces vanish separately:
11
Also this is a general phenomenon in BV theories of Chern-Simons type: The vanishing of the bound-
ary contributions from principal faces is ensured by the classical master equation. See for instance
[CMR17],[Cat19].

95
 a b a b a b
e

+ + =0
e e

e e
e
c d c d c d

Figure 3.3: The famous IHX relation, which holds whenever graphs are identical outside of
the dashed circle.

Lemma 3.4.5. Let S ⊂ V (Γ) have cardinality |S| ≥ 3. Then

Z
ωΓ = 0.
∂S CΓ (M )

Proof. This follows from the “vanishing Lemmata” by Kontsevich ([Kon94]), but we briefly
repeat the argument. The dimension of C̃S (R3 ) is 3|S| − 4. Consider the edge subgraph ΓS
on the vertices S ⊂ V (Γ) (i.e. it contains all edges between vertices of γ but no leaves).
Suppose all vertices in S are 3-valent. Then the corresponding propagators multiply to a 3|S|
form and after integrating over the fiber we have a 4-form placed at the point of collapse of S
(the point labeled [S] in CΓ/S ), which is zero for dimensional reasons. Thus the contribution
vanishes unless ΓS has a vertex of valence 2 or less. If a vertex has valence 0 or 1, integrating
over that vertex yields 0 (again for degree reasons). So the only remaining case is when ΓS
has a vertex of valence 2. In that case, we consider the integral over that vertex only, it
looks like Z
η(x, y)η(y, z).
y∈R3

On C̃S (R3 ), η is translation invariant, and the involution y 7→ x + z − y sends η(x, y)η(y, z)
to minus itself. We conclude that the integral is zero.

96
3.5 Gauge invariance and the framing anomaly

In the language of BV formalism, what we expect to be gauge invariant is the BV cohomology

A0
class of the partition function ZCS (a; g). More precisely, from Theorem 2.4.7 we expect that
if the vary the gauge-fixing Lagrangian smoothly, the partition function changes by a ∆Y -
exact term. In this section we will see that this is not the case, but that there is a partial
remedy to this. This is a well-known fact whose appearance in the literature will be discussed
in more detail in Subsection 3.5.3 below.

3.5.1 Dependence of Zf ree on the gauge-fixing metric

Since we only consider gauge-fixings coming from Riemannian metrics, we will consider gauge
fixing Lagrangians Lgt induced by a smooth family gt of Riemannian metrics. Surprisingly,
not even the “free” part of the partition function
iπ
Zf ree = τ (M, A0 )1/2 e 4 ψ(A0 ,g)

is gauge invariant. To compute the dependence on the gauge it is necessary to introduce a

framing and the gravitational Chern-Simons invariant:

Definition 3.5.1. A framing of a three-manifold M is a vector bundle isomorphism f : M ×

R3 → T M , i.e. a trivialization of the tangent bundle. We demand that, if M is oriented
then f is orientation-preserving.

The datum of a framing is equivalent to a section of the frame bundle sf : M → Fr(T M )

(the section s is simply given by sf (x) = (f (e1 ), f (e2 ), f (e3 )) where ei is the canonical basis
of R3 ). Using this section we can define the gravitational Chern-Simons invariant of g and
f:

Definition 3.5.2. Let g be a Riemannian metric on M and f a framing of M . Denote

ΩLC ∈ Ω1 (Fr(T M ), so(R3 )) the connection form of the Levi-Civita connection of the metric
g. Then we define Z
grav
SCS (g, f ) := s∗f cs(ΩLC ) (3.53)
M
with the Chern-Simons form defined as in Definition 1.2.4.

97
We then have the following proposition.

Proposition 3.5.3. For any framing f , the quantity

iπ i dim G 1 grav
ψ(A0 , g) + S (g, f ) (3.54)
4 24 2π CS
is independent of the metric g.

The proof uses the Atiyah-Patodi-Singer index theorem as in [Wit89] and goes well beyond
the scope of these notes.12 We conclude that if we rescale the free partition function by a
phase factor,  
i dim G 1 grav
Zf0 ree = exp S (g, f ) Zf ree (3.55)
24 2π CS
then Zf0 ree is gauge invariant but depends on the framing.

Dependence on the framing

grav
The dependence on the framing is controllable. First of all, SCS (g, f ) depends only on
the homotopy class of f . Homotopy classes of framings are slightly subtle: If we fix one
framing f , then any other framing is related to it by a map γ : M → SO(3), these maps are
distinguished by their degree deg γ, which is an integer, and an element c(γ) ∈ H 1 (M, Z2 ).13
If we modify the framing by a map γ : M → SO(3) of degree n with c(γ) = 0, then the
gravitational Chern-Simons invariant changes by
1 grav 1 grav
SCS (M, f · γ) = S (M, f ) + 2πn (3.56)
2π 2π CS
and thus a change of framing changes the rescaled partition function by a phase factor of
 
0 0 2πin dim G
Zf ree (f · γ) = Zf ree (f ) exp . (3.57)
24
12
A thorough account of the APS theorem, which also explains (3.54), is Melrose [Mel93] - almost 400
pages!
13
The degree can be computed as deg γ = M γ ∗ ω, where ω is a normalized volume form on SO(3), c(γ)
R

is the pullback γ ∗ α, where α ∈ Ω1 (M ) is a generator of H 1 (SO(3), Z2 ) (i.e. α is closed and for a generator
R
τ of π1 (SO(3)) = Z2 , τ α = 1.)

98
3.5.2 Gauge dependence of Zpert

The appearance of the gravitational Chern-Simons action is a foreshadowing of what happens

at higher order corrections. We present here only a brief version, and again refer to the
literature for technical details. To analyze the time derivative of the higher order correction
we first need to know what happens to the propagator and the residual fields.
Lemma 3.5.4. Let gt be a smooth family of Riemannian metrics and denote ηt the propagator
i
defined by gt , χi(k),t a basis of harmonic k-forms for gt . Then there are ξ(k) ∈ Ωk−1 (M, g) and
λ ∈ Ω1 (M × M, g ⊗ g) such that
d i
χi(k),t = dA0 ξ(k) (3.58)
dt t=0
d X
ηt = dA0 λ + π1∗ ξ(k)
i
π2∗ χi(k) + π1∗ χi(k) π2∗ ξ(k)
i
(3.59)
dt t=0 i,k

(3.60)

Proof. The first equation follows immediately from the fact that the cohomology class of
χi(k),t is independent of t. For the second statement, see [CM08].

Stokes’ theorem, again

t
This allows us to apply Stokes’ theorem in a similar fashion to the above. Denote by Zpert
the perturbative part of the partition function defined using the metric gt , then we have
d d XZ XZ d
t
Zpert = ω̃Γ = ω̃Γ
dt t=0 dt t=0 Γ CΓ (M ) Γ CΓ (M ) dt t=0
X X Z X Z
e
= ω̃Γ + ω̃Γl
Γ e∈E(Γ) CΓ (M ) l∈L(Γ) CΓ (M )

where now we denote ω̃ e the form obtained by placing η̇ at the edge e and ω̃ l the form
P (k) i
obtained by placing ȧ = i,k zi dξ(k) =: dA0 b at the leaf l. In this sum, we integrate all
terms of the form dλ or dξ by parts. Up to boundary terms, this leaves us with terms where
dA0 is applied to propagators. An argument similar to Proposition 3.4.3 shows that these
terms sum up to ∆X, where X is given by
XZ XZ
0
X= e
(ω̃Γem ) + (ω̃Γl lm )0 . (3.61)
Γem CΓ (M ) CΓ (M )
Γlm

99
Here Γem (Γlm ) runs over graphs with a marked edge (leaf), Γ denotes the graph with the
marking forgotten and ω̃Γe em (ω̃Γe em ) denotes the form obtained from the usual Feynman rules
but putting λ (resp. b) at the marked edge (resp. leaf). The only slight difference are the
terms of the form χξ in η: these arise when ∆Y eats an a-b pair of residual fields (instead of
the usual a-a-pair).

The boundary terms

For the boundary terms, one performs an analysis similar to the above. Again, summing
over all graphs boundary faces corresponding to |S| = 2 cancel out by the IHX relation. For
the “hidden faces”, the only difference is the degree count (since λ has degree 1) in the case
where an entire connected component of a graph collapses that contains no residual fields
(all vertices of the edge subgraph ΓS on S ⊂ V (Γ) are trivalent) and such that the marked
edge is contained in ΓS . In these cases the degree count now says that we obtain a 3-form
at the point of collapse. One can compute the integral of this 3-form and show that is given
by a numeric coefficient φ(Γ, g), which depends only on the graph Γ and the Lie algebra g
times the derivative of the gravitational Chern-Simons action, for any framing f :
Z Z !
d grav
(ωΓe )00 = φ(ΓS , g) SCS (gt , f ) (3.62)
∂S Γ CΓ−ΓS (M ) dt t=0

Rescaling of the entire partition function

Summing over all graphs, we see that we obtain

d X d grav
Zpert = ∆X + Zpert × φ(Γ, g) SCS (gt , f ). (3.63)
dt t=0 Γconnected
dt t=0

We have thus arrived at the following result.

Theorem 3.5.5. There exists a power series φ(~) of the form

X (−i~)χ(Γ)
φg (~) = φ0 + φ(Γ, g) (3.64)
Γ
|Aut(Γ)|

such that for every framing f , the BV cohomology class of the rescaled partition function
grav
A0 0 A0
(ZCS ) (a, g, f ) = ZCS (a, g)eiφg (~)SCS (g,f )
(3.65)

100
 2π dim G
does not depend on the metric g. Moreover φ0 = 24
.

In particular, the BV cohomology class of the rescaled partition function is an invariant of the
framed 3-manifold M . Moreover, under a change of framing the gravitational Chern-Simons
action changes according to Equation 3.56.
Remark 3.5.6. For degree reasons, φg can have only components in even degrees of ~. One
can show (see e.g. [CM08]) that the coefficient of ~2 (i.e. the two-loop anomaly) is given
by C2 (g)/48. It was conjectured by Axelrod and Singer [APW91],[AS94] that all higher
coefficients vanish. This conjecture is still open.
Remark 3.5.7 (Dependence on the reference flat connection). It is also possible to prove - see
[MW21] - that the BV cohomology class of the partition function depends only on the class
[A0 ] in the moduli space of flat connections. Moreover, at smooth points one can analyze
how the partition function varies as a function of [A0 ]. Ultimately, one would like to extract
a number-valued partition function - along the lines of [CMR17, Appendix F] - and compare
it to the asymptotics of the Reshetikhin-Turaev invariant [RT91]. This will be the focus of
future research.

3.5.3 The framing anomaly in the literaure

The framing anomaly in Chern-Simons theory was observed in various places, starting with
Witten [Wit89]. In the perturbative setting, this anomaly was discussed [AS91],[AS94] and
later in [BC98],[BC99], [Cat99],[CM08]. In the Kontsevich-Kuperberg-Thurston-Lescop ap-
proach [Kon94],[KT99],[Les04a],[Les04b] a framing is used to define the propagator, and the
resulting perturbative series depends on it.
In contrast, the Reshetikhin-Turaev invariants [RT91] do not exhibit any dependence on
a framing. This initially was the source of some confusion, since there is no canonical
choice of framing for a 3-manifold that could have been implicit in the construction of the
Reshetikhin-Turaev invariant. A possible explanation was offered by Atiyah, who showed in
[Ati90] that 3-manifolds admit a canonical 2-framing. He argued that Chern-Simons invari-
ants should more naturally be considered as invariants of 2-framed manifolds (and that the
Reshetikhin-Turaev invariants correspond to the Chern-Simons invariants in the canonical
2-framing). However, for perturbative Chern-Simons invariants the role of 2-framings has

101
yet to be understood precisely.

102
Acknowledgements

This review article grew from a series of lectures the author gave at the KMPG at Humboldt-
Universität Berlin in the summer of 2019 to an audience consisting of graduate students and
researchers.
It is aimed at students at the early graduate level - the prerequisite are basic knowledge of
manifolds and abstract algebra - with the explicit goal to introduce them to a topic at the
forefront of current research. The goal of these notes is to be as self-contained as possible,
but to not overload students or interested researchers with unnecessary details. A necessary
trade-off is that the notes often only scratch the surface of much deeper topics, choosing to
refer to the literature for a more detailed discussion for those who are interested. During his
time at HU, the author was supported by a Dirichlet Postdoctoral fellowship of the Berlin
Mathematical school. The author is very grateful for the hospitality shown by the math
department of HU and the BMS, especially by the group of Dirk Kreimer. The author is
also very grateful for the interest and criticism the attendees showed during the lectures, and
for the many engaging discussions. The author would like to thank D. Kreimer, M. Berghoff
and D. Prinz in particular for comments on the lectures and the notes.
The author would also like to thank A. Cattaneo and P. Mnev for many enlightening dis-
cussions about this subject. During completion of the article the author was also supported
by the SNF [Link] grant P2ZHP2 184083. Finally, the author wants to thank the
referee for many helpful suggestions, spotting countless errors big and small, and helping to
make this a much better text overall.

103
Bibliography

[AF17] D. Ayala and J. Francis. The cobordism hypothesis. 2017. arXiv: 1705.02240v2
[[Link]].
[Ale+97] M. Alexandrov, A. Schwarz, O. Zaboronsky, and M. Kontsevich. “The geometry
of the master equation and topological quantum field theory”. In: Internat. J.
Modern Phys. A 12.7 (1997), pp. 1405–1429.
[Ans94] D. Anselmi. “Removal of divergences with the Batalin-Vilkovisky formalism”.
In: Classical and Quantum Gravity 11.9 (1994), pp. 2181–2204.
[APS73] M. F. Atiyah, V. K. Patodi, and I. M. Singer. “Spectral Asymmetry and Rie-
mannian Geometry”. In: Bulletin of the London Mathematical Society 5.2 (1973),
pp. 229–234.
[APS75] M. F. Atiyah, V. K. Patodi, and I. M. Singer. “Spectral asymmetry and Rieman-
nian Geometry. I”. In: Mathematical Proceedings of the Cambridge Philosophical
Society 77.1 (1975), pp. 43–69.
[APW91] S. Axelrod, S. D. Pietra, and E. Witten. “Geometric quantization of Chern-
Simons gauge theory”. In: Journal of Differential Geometry 33.3 (1991), pp. 787–
902.
[AS91] S. Axelrod and I. M. Singer. “Chern-Simons perturbation theory”. In: Differ-
ential geometric methods in theoretical physics, Proceedings, New York. Vol. 1.
1991, pp. 3–45.
[AS94] S. Axelrod and I. M. Singer. “Chern-Simons perturbation theory. II”. In: J.
Differ. Geom. 39.1 (1994), pp. 173–213.

104
[Ati88] M. Atiyah. “Topological quantum field theories”. In: [Link]. IHÉS 68.1
(1988), pp. 175–186.
[Ati90] M. Atiyah. “On framings of 3-manifolds”. In: Topology 29.1 (1990), pp. 1–7.
[AU15] J. E. Andersen and K. Ueno. “Construction of the Witten-Reshetikhin-Turaev
TQFT from conformal field theory”. In: Invent. Math. 201.2 (2015), pp. 519–559.
[Bar91] D. Bar-Natan. “Perturbative aspects of the Chern-Simons topological quantum
field theory”. PhD thesis. Princeton U., 1991.
[Bau14] H. Baum. Eichfeldtheorie. Springer Berlin Heidelberg, 2014. 396 pp.
[BC98] R. Bott and A. S. Cattaneo. “Integral invariants of 3-manifolds”. In: J. Differ.
Geom. 48.1 (1998), pp. 91–133.
[BC99] R. Bott and A. S. Cattaneo. “Integral invariants of 3-manifolds II”. In: J. Differ.
Geom. 53.1 (1999), pp. 1–13.
[BD13] T. Brocker and T. tom Dieck. Representations of Compact Lie Groups. 2013.
[BD95] J. C. Baez and J. Dolan. “Higher-dimensional algebra and topological quantum
field theory”. In: J. Math. Phys. 36.11 (1995), pp. 6073–6105.
[Ber66] F. A. Berezin. The Method of Second Quantization (Monographs and Textbooks
in Pure and Applied Physics, Vol. 24). Academic Press, 1966.
[BGV03] N. Berline, E. Getzler, and M. Vergne. Heat Kernels and Dirac Operators (Grundlehren
der mathematischen Wissenschaften). Springer, 2003.
[BRS75] C. Becchi, A. Rouet, and R. Stora. “Renormalization of the abelian Higgs-Kibble
model”. In: Commun. Math. Phys. 42.2 (1975), pp. 127–162.
[BRS76] C. Becchi, A. Rouet, and R. Stora. “Renormalization of Gauge Theories”. In:
Ann. Phys. 98 (1976), pp. 287–321.
[BT94] R. Bott and C. Taubes. “On the self-linking of knots”. In: J. Math. Phys. 35.10
(1994), pp. 5247–5287.
[BV81] I. Batalin and G. Vilkovisky. “Gauge algebra and quantization”. In: Phys. Lett.
B 102.1 (1981), pp. 27–31.
[BV83] I. A. Batalin and G. A. Vilkovisky. “Quantization of gauge theories with linearly
dependent generators”. In: Phys. Rev. D 28.10 (1983), pp. 2567–2582.

105
[Cat+95] A. Cattaneo, B. Keller, C. Torossian, and A. Bruguières. Déformation, Quan-
tification, Théorie de Lie, Panoramas et Syntheses 20. Société Mathématique de
France, 1995.
[Cat19] A. Cattaneo. On the BV formalism. 2019. url: [Link]
cattaneo/bv_catt.pdf.
[Cat99] A. Cattaneo. “Configuration space integrals and invariants for 3-manifolds and
knots”. In: Low Dimensional Topology. Ed. by H. Nencka. Cont. Math. 233. 1999,
pp. 153–165.
[CF00] A. S. Cattaneo and G. Felder. “A path integral approach to the Kontsevich
quantization formula”. In: Commun. Math. Phys. 212 (2000), pp. 591–611.
[Che77] J. Cheeger. “Analytic torsion and Reidemeister torsion”. In: Proceedings of the
National Academy of Sciences 74.7 (1977), pp. 2651–2654.
[Che79] J. Cheeger. “Analytic Torsion and The Heat Equation”. In: The Annals of Math-
ematics 109.2 (1979), p. 259.
[CM08] A. S. Cattaneo and P. Mnev. “Remarks on Chern-Simons Invariants”. In: Com-
mun. Math. Phys. 293 (2008), pp. 803–836.
[CMR14] A. S. Cattaneo, P. Mnev, and N. Reshetikhin. “Classical BV Theories on Mani-
folds with Boundary”. In: Commun. Math. Phys. 332.2 (2014), pp. 535–603.
[CMR17] A. S. Cattaneo, P. Mnev, and N. Reshetikhin. “Perturbative Quantum Gauge
Theories on Manifolds with Boundary”. In: Commun. Math. Phys. 357.2 (2017),
pp. 631–730.
[CMW17] A. S. Cattaneo, P. Mnev, and K. Wernli. “Split Chern–Simons Theory in the BV-
BFV Formalism”. In: Quantization, Geometry and Noncommutative Structures
in Mathematics and Physics. Springer International Publishing, 2017, pp. 293–
324.
[CMW18] A. S. Cattaneo, P. Mnev, and K. Wernli. Theta Invariants of lens spaces via
the BV-BFV formalism. 2018. arXiv: [Link]
[[Link]].
[Cos11] K. Costello. Renormalization and Effective Field Theory. Mathematical Surveys
and Monographs 170. American Mathematical Society, 2011.

106
[Cra04] M. Crainic. On the perturbation lemma, and deformations. 2004. arXiv: http:
//[Link]/abs/math/0403266v1 [[Link]].
[CS11] A. S. Cattaneo and F. Schätz. “Introduction to Supergeometry”. In: Reviews in
Mathematical Physics 23.06 (2011), pp. 669–690.
[CS74] S.-S. Chern and J. Simons. “Characteristic Forms and Geometric Invariants”.
In: Ann. of Math. (2) 99.1 (1974), pp. 48–69.
[Fey42] R. P. Feynman. “The Principle of Least Action in Quantum Mechanics”. Thesis
(Ph.D.) Princeton, NJ, USA: Department of Physics, Princeton University, 1942.
[Fey49a] R. P. Feynman. “Space-Time Approach to Quantum Electrodynamics”. In: Phys.
Rev. 76 (6 1949), pp. 769–789.
[Fey49b] R. P. Feynman. “The Theory of Positrons”. In: Phys. Rev. 76 (6 1949), pp. 749–
759.
[Fey50] R. P. Feynman. “Mathematical Formulation of the Quantum Theory of Electro-
magnetic Interaction”. In: Phys. Rev. 80 (3 1950), pp. 440–457.
[FH16] D. S. Freed and M. J. Hopkins. Reflection positivity and invertible topological
phases. 2016. arXiv: 1604.06527v3 [hep-th].
[Fio03] D. Fiorenza. “An introduction to the Batalin-Vilkovisky formalism”. In: Comptes
Rendus des Rencontres Mathematiques de Glanon, Edition 2003. 2003.
[FK14] G. Felder and D. Kazhdan. “The classical master equation”. In: Perspectives in
representation theory : a conference in honor of Igor Frenkel’s 60th birthday.
Ed. by P. Etingof, M. Khovanov, and A. Savage. Contemporary Mathematics.
American Mathematical Society, 2014, pp. 79–137.
[FK89] J. Fröhlich and C. King. “The Chern-Simons theory and knot polynomials”. In:
Commun. Math. Phys. 126.1 (1989), pp. 167–199.
[FQ93] D. S. Freed and F. Quinn. “Chern-Simons theory with finite gauge group”. In:
Communications in Mathematical Physics 156.3 (1993), pp. 435–472.
[Fre02] D. S. Freed. “Classical Chern-Simons Theory, Part 2”. In: Houston J. Math. 28
(4 2002), pp. 293–.
[Fre12] D. S. Freed. “The Cobordism hypothesis”. In: Bulletin of the American Mathe-
matical Society 50.1 (2012), pp. 57–92.

107
[Fre95] D. S. Freed. “Classical Chern-Simons Theory, 1”. In: Advances in Mathematics
113.2 (1995), pp. 237–303.
[GJ87] J. Glimm and A. Jaffe. Quantum Physics. Springer New York, 1987.
[GMM89] E. Guadagnini, M. Martellini, and M. Mintchev. “Perturbative aspects of the
Chern-Simons field theory”. In: Phys. Lett. B 227.1 (1989), pp. 111–117.
[Hol08] S. Hollands. “RENORMALIZED QUANTUM YANG–MILLS FIELDS IN CURVED
SPACETIME”. In: Reviews in Mathematical Physics 20.09 (2008), pp. 1033–
1172.
[HT94] M. Henneaux and C. Teitelboim. Quantization of Gauge Systems. Princeton
University Press, 1994.
[Hus93] D. Husemöller. Fibre Bundles. Springer New York, 1993. 380 pp.
[Joh10] T. Johnson-Freyd. “On the coordinate (in)dependence of the formal path inte-
gral”. In: (2010).
[Jon85] V. F. R. Jones. “A polynomial invariant for knots via von Neumann algebras”.
In: Bulletin of the American Mathematical Society 12.1 (1985), pp. 103–112.
[Kap10] A. Kapustin. “Topological field theory, higher categories, and their applications”.
In: Proceedings of the International Congress of Mathematicians 2010 (ICM
2010) (In 4 Volumes) Vol. I: Plenary Lectures and Ceremonies Vols. II–IV:
Invited Lectures. World Scientific. 2010, pp. 2021–2043.
[Khu04] H. M. Khudaverdian. “Semidensities on Odd Symplectic Supermanifolds”. In:
Commun. Math. Phys. 247.2 (2004), pp. 353–390.
[KN96] S. Kobayashi and K. Nomizu. Foundations of Differential Geometry. Wiley-
Interscience, 1996. 348 pp.
[Kna96] A. W. Knapp. Lie Groups Beyond an Introduction. Birkhäuser Boston, 1996.
[Koc03] J. Kock. Frobenius Algebras and 2D Topological Quantum Field Theories. Cam-
bridge University Press, 2003.
[Kon93] M. Kontsevich. “Vassiliev’s knot invariants”. In: Adv. in Sov. Math 16.2 (1993),
pp. 137–150.

108
[Kon94] M. Kontsevich. “Feynman Diagrams and Low-Dimensional Topology”. English.
In: First European Congress of Mathematics Paris, July 6–10, 1992. Ed. by A.
Joseph, F. Mignot, F. Murat, B. Prum, and R. Rentschler. Vol. 120. Progress in
Mathematics. Birkhäuser Basel, 1994, pp. 97–121.
[Kre08] D. Kreimer. “A remark on quantum gravity”. In: Annals of Physics 323.1 (2008),
pp. 49–60.
[KT17] A. Kapustin and A. Turzillo. “Equivariant topological quantum field theory and
symmetry protected topological phases”. In: J. High Energy Phys. 2017.3 (2017).
[KT99] G. Kuperberg and D. P. Thurston. Perturbative 3-manifold invariants by cut-
and-paste topology. 1999. eprint: math/9912167.
[Les04a] C. Lescop. On the Kontsevich-Kuperberg-Thurston construction of a configuration-
space invariant for rational homology 3-spheres. 2004. arXiv: math/0411088v1
[[Link]].
[Les04b] C. Lescop. Splitting formulae for the Kontsevich-Kuperberg-Thurston invariant
of rational homology 3-spheres. 2004. arXiv: math/0411431v1 [[Link]].
[Lur09] J. Lurie. “On the Classification of Topological Field Theories”. In: arXiv preprint
(2009), pp. 129–280.
[Mel93] R. Melrose. The Atiyah-Patodi-Singer Index Theorem. A K Peters/CRC Press,
1993.
[Mne17] P. Mnev. Lectures on Batalin-Vilkovisky formalism and its applications in topo-
logical quantum field theory. 2017. arXiv: 1707.08096 [math-ph].
[Mül78] W. Müller. “Analytic torsion and R-torsion of Riemannian manifolds”. In: Ad-
vances in Mathematics 28.3 (1978), pp. 233–305.
[MW21] P. Mnev and K. Wernli. “On the dependence of the perturbative Chern-Simons
partition function on the reference flat connection. In preparation”. 2021.
[Nic03] L. I. Nicolaescu. The Reidemeister Torsion of 3-Manifolds. Walter de Gruyter,
2003.
[Nic07] L. I. Nicolaescu. Lectures on the Geometry of Manifold. World Scientific Pub-
lishing Company, 2007. 608 pp.

109
[Pol05] M. Polyak. “Feynman diagrams for pedestrians and mathematicians”. In: Proc.
Symp. Pure Math. 73 (2005), pp. 15–42.
[Pri18] D. Prinz. Algebraic Structures in the Coupling of Gravity to Gauge Theories.
2018. arXiv: [Link] [hep-th].
[PS95] M. E. Peskin and D. V. Schroeder. An Introduction To Quantum Field Theory.
Taylor & Francis Inc, 1995. 866 pp.
[Rei35] K. Reidemeister. “Homotopieringe und Linsenräume”. In: Abhandlungen aus
dem Mathematischen Seminar der Universität Hamburg 11.1 (1935), pp. 102–
109.
[Res10] N. Reshetikhin. “Lectures on Quantization of Gauge Systems”. In: New Paths
Towards Quantum Gravity. Springer Berlin Heidelberg, 2010, pp. 125–190.
[Rha84] G. D. Rham. Differentiable Manifolds. Springer Verlag, 1984.
[RS71] D. Ray and I. Singer. “R-Torsion and the Laplacian on Riemannian manifolds”.
In: Advances in Mathematics 7.2 (1971), pp. 145–210.
[RT91] N. Reshetikhin and V. G. Turaev. “Invariants of 3-manifolds via link polynomials
and quantum groups”. In: Invent. Math. 103.1 (1991), pp. 547–597.
[Sch78] A. Schwarz. “The partition function of degenerate quadratic functional and Ray-
Singer invariants”. English. In: Lett. Math. Phys. 2.3 (1978), pp. 247–252.
[Sch93] A. Schwarz. “Geometry of Batalin-Vilkovisky quantization”. In: Commun. Math.
Phys. 155.2 (1993), pp. 249–260.
[Seg88] G. B. Segal. “The definition of conformal field theory”. In: Differential geometri-
cal methods in theoretical physics (Como, 1987). Vol. 250. NATO Adv. Sci. Inst.
Ser. C Math. Phys. Sci. Dordrecht: Kluwer Acad. Publ., 1988, pp. 165–171.
[Šev06] P. Ševera. “On the Origin of the BV Operator on Odd Symplectic Supermani-
folds”. In: Lett. Math. Phys. 78.1 (2006), pp. 55–59.
[Sin03] D. Sinha. Manifold theoretic compactifications of configuration spaces. 2003.
arXiv: [Link] [[Link]].
[Tau11] C. H. Taubes. Differential Geometry: Bundles, Connections, Metrics and Cur-
vature. Oxford University Press, 2011. 312 pp.

110
[Tur94] V. G. Turaev. “Axioms for topological quantum field theories”. en. In: Annales de
la Faculté des sciences de Toulouse : Mathématiques Ser. 6, 3.1 (1994), pp. 135–
152.
[Tyu76] I. V. Tyutin. “Gauge Invariance in Field Theory and Statistical Physics in Oper-
ator Formalism”. In: Preprints of P.N. Lebedev Physical Institute, No. 39 (1976).
[Var04] V. S. Varadarajan. Supersymmetry for Mathematicians: An Introduction. Vol. 11.
Courant Lecture Notes. Courant Institute of Mathematical Sciences at New York
University, 2004.
[Wal92] K. Walker. An Extension of Casson’s Invariant. (Am-126), Volume 126. PRINCE-
TON UNIV PR, 1992. 150 pp.
[Wei05] S. Weinberg. The Quantum Theory of Fields. Cambridge University Press, 2005.
510 pp.
[Wit89] E. Witten. “Quantum field theory and the Jones polynomial”. In: Commun.
Math. Phys. 121.3 (1989), pp. 351–399.

111